ing, that persons use to describe the affective quality of .... a vector originating from the center of the circle. The representation of affective quality illustrated in ...
Journal of Personality and Social Psychology 1980, Vol. 38, No. 2, 311-322
A Description of the Affective Quality Attributed to Environments James A. Russell and Geraldine Pratt University of British Columbia, Vancouver, Canada The meaning that persons attribute to environments is divided into perceptualcognitive meaning and affective meaning. Affective meaning is then conceptualized as a two-dimensional bipolar space that can be denned by eight variables falling in the following circular order around the perimeter: pleasant (arbitrarily set at 0°), exciting (45°), arousing (90°), distressing (135°), unpleasant (180°), gloomy (225°), sleepy (270°), and relaxing (315°, which is thus 45° from pleasant). Alternatively, the same space can be denned by two orthogonal bipolar dimensions of pleasant-unpleasant and arousing-sleepy—or equally well by exciting-gloomy and distressing-relaxing. Reliable verbal scales for these eight variables are developed and shown to approximate the proposed theoretical structure.
Most of us would acknowledge the pervasive if subtle influence of affective responses on person-environment interactions. The English language provides hundreds of words, such as lively, boring, disgusting, and relaxing, that persons use to describe the affective quality of places. In spite of, or perhaps because of, this large array of affective descriptors, environmental psychologists have focused their attention on only a few and then only one at a time (such as degree of comfort, annoyance, pleasantness, or psychological stress), implicitly expressing pessimism about the possibility of including in their research or theorizing anything approaching a complete description and assessment of the affective quality attributed to places. Our thesis is that a very simple conceptualization can encompass the diverse affective concepts applied to molar physical environments. Theorists from widely differing vantage This study was supported by a research grant from the Canada Council to the first author. We are grateful to Nicole Clement and Klaus Schroeder for their help in carrying out this study and to Jerry S. Wiggins for helpful comments on an earlier draft of this article. Requests for reprints should be sent to James A. Russell, Department of Psychology, University of British Columbia, Vancouver, British Columbia, Canada V6T 1WS.
points have included affective concepts in their characterizations of environment-behavior relations. Ittelson (1973), a perceptionist, said, "The first level of response to the environment is affective. The direct emotional impact of the situation . . . very generally governs the directions taken by subsequent relations with the environment" (p. 16). Freedman (197S), a social psychologist, proposed more specifically that the psychological effects of a crowded environment are mediated by the pleasantness of the other factors in that environment. Other environmental psychologists have employed the concept of psychological stress (e.g., Glass & Singer, 1972; Stokols, 1972), the related concept of overload (e.g., Milgram, 1970), or other concepts with a clear affective component: arousal, environmental quality, comfort, annoyance, aesthetically pleasing quality, and so on. The present study continues a line of research aimed at providing a general description of affect (e.g., Russell, 1978, 1979) and of the affective quality attributed to environments (e.g., Mehrabian & Russell, 1974). Affect is defined here as emotion expressed in language, and affective quality of a molar physical environment (or more simply expressed, a place), as the emotion-inducing
Copyright 1980 by the American Psychological Association, Inc. 0022-3514/80/3802-0311$00.75
311
312
JAMES A. RUSSELL AND GERALDINE PRATT
Forceful Hectic
A c•t i v e
Exciting •
Frightening • Harsh
Majestic Festive
• Frustrating
Enjoyable Ugly
• Beautiful
Disgusting Unstimulating
Comfortable •
Depressing Boring
«
Insignificant * Peaceful
Desolate Sleepy
Figure 1. Two principal components of 21 clusters of adjectives descriptive of the affective quality of places.
quality that persons verbally attribute to that place. The goal of this research is thus to provide a conceptual structure that defines the meaning of terms within the domain of the English language that persons commonly use to describe the emotional quality of environments. In one previous study (Russell, Ward, & Pratt, Note 1), that domain was represented by 105 commonly used adjectives. These adjectives were factor analyzed, and based on theoretical and empirical considerations, two factors were found adequate to characterize their affective meaning. (Subsequent factors were interpreted as more perceptual/cognitive than affective in nature.) These results are illustrated in Figure 1, in which 21 representative adjectives were placed in the obtained two-dimensional space. It is easy to label the axes. Dimension 1 ranges from unpleasant to pleasant, and Dimension 2 ranges from sleepy to arousing. What is more important to note, however, is that the various terms do not
cluster about these axes (do not form a simple structure), but are placed meaningfully throughout the space. Exciting, for instance, should not be viewed as meaning either pleasant or arousing alone, but must be seen as meaning the combination of pleasant and arousing. Similarly, peaceful is the combination of pleasant and unarousing; boring is the combination of unpleasant and unarousing; frightening is the combination of unpleasant and arousing. In Figure 2 we propose a geometrical representation of various properties of this domain of terms. The existence of semantic opposites (antonyms such as unpleasant vs. pleasant or exciting vs. gloomy), for example, is represented by the dimensions' being bipolar. The proposed model illustrated in Figure 2 is a two-dimensional* bipolar space 1 More precisely, two dimensions are assumed sufficient to represent the emotional, but not perceptual-cognitive, component of the meaning of these
AFFECTIVE QUALITY OF PLACES
313
in which eight terms are placed approximately 45° apart in a circular order. The space is meant to be continuous, however, so that the affective quality attributed to a particular environment could fall at any point in the terms. Our use of only two dimensions to represent even the emotional component of affect requires some further comment, since earlier work (Mehrabian & Russell, 1974) proposed three dimensions. Three dimensions had been suggested by the semantic differential evidence, based on Osgood's (1969) interpretation of evaluation, activity, and potency as affective. Mehrabian and Russell then proposed the following three analogous concepts to describe affect (renamed to better reflect this interpretation): pleasure, arousal, and dominance. Subsequent work indicated a more complex situation, however. In the Russell et al. (Note 1) study, a measure of dominance accounted for only a trivial proportion of variance in the obtained factor structure of the affective quality attributed to places and, contrary to expectation, was not equivalent to the potency factor obtained. Separately, multidimensional scaling studies of affect terms (e.g., Russell, 1978) indicated that besides dominance, there are a number of dimensions beyond pleasure and arousal contained in the meaning of affect terms. These were dimensions such as locus of causation, importance of the emotion, and locus of control. Whereas pleasure and arousal each account for large proportions of variance in the meaning of affect terms, each dimension beyond these two accounted for only a tiny proportion. More importantly, these secondary dimensions became more and more clearly interpretable as cognitive rather than emotional in nature. The secondary dimensions thus appear to be aspects of the cognitive appraisal system that has been suggested for emotions. Indeed, dimensions such as locus of causation and importance are general informationprocessing concepts applicable to a wide range of phenomena and not unique to emotions. Because the secondary dimensions account for quite small amounts of variance, they are difficult to uncover, but by extrapolation, one would have to anticipate the discovery of even more such cognitive dimensions. The question then becomes where to draw the distinction between the emotional and the cognitive dimensions underlying the meaning of affect terms. Our view is that two dimensions—pleasure-displeasure and degree of arousal—denote the internal emotional state per se. All other dimensions are cognitive in that they denote beliefs about antecedents, consequences, or other such properties of the emotional state. The present article focuses on a description of the emotional component of affect, but further exploration and conceptualization of the cognitive component of affect is, of course, just as important. However, we do not expect findings in this latter area to necessitate altering our description of the emotional component of affect.
Figure 2. Circular ordering descriptors.
of
eight
affective
space. Any affect term can be represented as a vector originating from the center of the circle. The representation of affective quality illustrated in Figure 2 has the advantage of offering a network of testable propositions in which all variables are interrelated. Thus, if the eight variables shown were exactly 45° apart and were measured without error, correlations among them should be perfectly accounted for by two underlying factors. Each variable also has a bipolar opposite with which it should correlate —1.00. More generally, each variable should correlate in the following manner with the seven others: .707 with the variables 45° away, .00 with the one 90° away, —.707 with the one 135° away, -1.00 with the one 180° away, —.707 with the one 225° away, .00 with the one 270° away, and .707 with the one 315° away. (The cosine between two vectors in a two-dimensional space gives their intercorrelation.)2 2 The proposed structure appears to fulfill the requirements of a circumplex, and Wiggins (1979) has recently discussed the advantages of using circumplex models to represent specific domains of variables. Wiggins, in turn, was influenced by writers such as Benjamin (1974), Foa and Foa (1974), Guttman (19S4), Leary (19S7), Lorr and McNair (1963), and Schaefer (1959). One common element in the approaches of these various writers is a belief in the advantages of representing a domain of highly interrelated variables within a geometrical space of small dimensionality, in which the variables are
314
JAMES A. RUSSELL AND GERALDINE PRATT
There is already empirical evidence concerning some of the predictions from this model—for example, that affective space is bipolar (Meddis, 1972; Russell, 1979), that the two principal axes are independent (Russell et al., Note 1), and that various terms can be denned as combinations of those two axes (Russell & Mehrabian, 1977). The purpose of the present study was to use this conceptualization as a guide in developing scales to assess the eight variables given in Figure 2. These scales, in turn, can be used both for further tests of the proposed conceptualization as well as in practical situations requiring assessment of the affective quality attributed to environments. Preliminary scales of these eight variables were thus first developed and then cross-validated. The scales were then further refined into eight final scales, and the relationships among these scales were compared with the theoretical structure. Use of the eight scales in basic and applied research is also discussed below.
vironments (e.g., old was excluded despite its affective connotation).3 Environments. Our procedure for sampling descriptions of environments was based on a definition of molar physical environments as including all those aspects of a person's physical surroundings that are directly perceivable. This definition is meant to exclude aspects of environments that are known but not perceived (such as an entire city) but to include a broad range of types of environments. With this diversity in mind, a list of 323 different environments in or around Vancouver, British Columbia, was constructed. The final sample of environments was indeed diverse, including places ranging from a wilderness area to a nightclub, from a bathroom to an airport, from an elevator to a nude beach. Subjects. We also sought a diversity of subjects. Thus, the experimenter went to the environment chosen and randomly selected one person already there, excluding only those who appeared to be under 16 years of age. Development of Preliminary Unipolar Scales That person was asked to fill out a short questionnaire concerning the environment. Data Pool (Experimenters reported very few cases of Data for preliminary scale construction anyone refusing.) For the more private and consisted of 323 cases, each of a separate remote environments, subjects were recruited subject describing a separate environment on a first and sent to the environment; this occurred 1 OS-item list of affect-denoting adjectives. in appoximately SO of the 323 cases. Each These data have already been described in subject described only one environment (dedetail elsewhere (Russell et al., Note 1) and fined as including anything he or she could will only be described briefly here. perceive without moving much) by rating the Item pool. A reasonably comprehensive descriptive accuracy of each of the 105 adsample of affective adjectives was compiled jectives on a scale from 1 (extremely inacfrom two sources: Craik's (1971) list of 204 curate) to 8 (extremely accurate). The subadjectives commonly used to describe land- ject then rated his or her personal emotional scapes and 585 words used by 52 college undergraduates asked to describe the affec3 The meaning attributed to environments contains tive quality of various man-made environ- both affective and perceptual-cognitive components, ments. The final 10S adjectives were selected with the two highly interrelated (on the assumption from the larger pool by excluding all words that the way in which something is responded to that (a) were used infrequently, (b) were not emotionally depends on how it is perceived and cognized). Since much confusion concerning environadjectives, and (c) contained perceptual/ mental meaning has arisen because these two aspects cognitive meaning relevant to physical en- of meaning were not distinguished (see Russell et al., arrayed in a circular order. Schaefer and Plutchik (1966) and Schlosberg (1954) have specifically proposed viewing emotions as forming a circular ordering. The present conceptualization can thus be seen as an attempt to propose a specific circumplex model for affective quality.
Note 1), we attempted to excude as much perceptual-cognitive meaning as possible from our sample of adjectives to focus on affective meaning. Were sufficient variance along dimensions of perceptualcognitive meaning included in our adjective sample the resulting factor structure would necessarily have reflected this fact.
AFFECTIVE QUALITY OF PLACES
reaction to the environment on Mehrabian and Russell's (1974) scales of pleasure, arousal, and dominance. Design. Wiggins (1973) distinguished between the internal and external structure of a set of terms. Internal structure of the language of affective quality is the implicational structure within that domain of terms. External structure concerns the application of the terms to some external stimulus. Our concern in the present study was with internal structure, and therefore the observations we obtained were sampled from the universe of descriptions of the affective quality. Our procedure was designed to obtain an adequate sample of such descriptions by seeking a diverse sample of subjects describing a diverse sample of environments. Although this is a powerful technique for obtaining data on internal structure, it does not allow us to examine what determined the descriptions (which would represent a concern with external structure), since subject and environment are confounded (each subject was in a different environment). Thus, the results do not bear on the extent to which a person's description of a particular place is accurate, or, more generally, on the relative contribution of the person, the environment, or their interaction in determining the description. To the extent that our sample of descriptions is representative (and this may be questionable, since observations were restricted to Vancouver), our procedure does allow a generalization of the results both to new subjects and to new environments. Thus for our purposes, this procedure was preferable to the more typical one of sampling from the universe of subjects, who then respond to a limited set of environments (with highly questionable generalizability to new environments), or the alternative procedure of sampling from the universe of environments, which are then presented to a limited set of subjects (with questionable generalizability to new subjects). Scale Construction The data pool was randomly divided into Sample A (N = 161), which was used for scale construction, and Sample B (N = 162),
315
which was used for cross-validation. Each subject's data were first ipsatized (equated with respect to mean and variance) to control a response style factor of individual differences in the use of the rating scale (Bentler, 1969, 1973). With the data from Sample A, two principal components were then extracted and rotated to approximate pleasing and arousing quality dimensions. All 105 adjectives were considered as possible candidates for the scales, since these items had already been selected as both affect denoting and commonly used. Items were then selected for the eight scales on the basis of appropriate loadings on the principal components, as dictated by the model illustrated in Figure 2 (i.e., moderately positive loadings on both components for exciting quality, high positive loading on the first component and near zero loading on the second component for pleasing quality, and so on). Items were also selected on the basis of appropriate correlations with Mehrabian and Russell's (1974) scales of pleasure and arousal. The eight groups of items selected in this way were then formed into preliminary sets. Items were then deleted from or added to each set, on the basis of their correlations with these preliminary sets, to maximize homogeneity of the scales with the minimum of items. Cross-Validation of Preliminary Scales Data from Sample B were then used to cross-validate the scale construction procedure so far. A principal components analysis of the eight scales (with unities on the diagonal) indicated that two components (with eigenvalues of 3.57 and 2.95) accounted for 81.57" °f ^e total variance. The next component accounted for a trivial proportion of variance (eigenvalue of .45), confirming our hypothesis that the eight scales form a twodimensional space. In Table 1 correlations of the eight scales with Mehrabian and Russell's (1974) scales of pleasure and arousal are given. These correlations showed the expected pattern and, hence, provided some evidence of external validity for the scales. As expected, the pleasure felt by a person correlated highly and
JAMES A. RUSSELL AND GERALDINE PRATT
316
Table 1 Reliabilities, Inter correlations, and Correlations With External Criteria for the Preliminary Eight Scales of Affective Quality
Scale
1. 2. 3. 4. 5. 6. 7. 8.
Arousing quality Exciting quality Pleasant quality Relaxing quality Sleepy quality Gloomy quality Unpleasant quality Distressing quality
Correlation with Mehrabian and Russell's (1974) scales
No. of items
a
5 5 5 5 3 5 1 5
.70 .84 .85 .90 .83 .86 .81 .78
Pleasure Arousal .17 .66 .67 .49 .09 -.66 -.77 -.40
.59 .44 .13 -.46 -.65 -.41 .03 .41
Intercorrelations among scales 2
3
4
.58 .06 -.49 .57 .09 .52
5
6
7
8
-.62 -.32 .03 .61
-.54 -.80 -.67 -.15 .29
-.08 -.63 -.74 -.60 -.23 .62
.42 -.08 -.53 -.76 -.54 .10 .55
Note. Statistics presented here are based on the cross-validation sample (N = 162), after each subject's data were ipsatized (equated with respect to mean and variance).
positively with the pleasing quality (.67) attributed to the environment. Pleasure also correlated highly and negatively with unpleasant quality ( — .77), near zero with both arousing (.17) and sleepy qualities (.09), moderately and positively with both exciting (.66) and relaxing qualities (.49), and moderately and negatively with both gloomy (— .6,6) and distressing qualities ( — .40). The predicted pattern for arousal can also be seen in Table 1. Additional results are shown in Table 1, but they will be discussed later in conjunction with similar results for the final scales. Development of Final Unipolar Scales Additional items were devised for the eight scales so that there were 10 potential items for each scale. These additional items were commonly used words, as close as possible in meaning to the cluster of items forming the preliminary scales. Dictionaries, thesauruses, and colleagues were consulted for new words. New data were then gathered with the resultant 80-item list for use in constructing a final set of scales. (Examples of items are given in Table 4.) The procedure for obtaining data was nearly identical to that used for gathering the preliminary data, except that the Mehrabian and Russell (1974) scales were omitted. The subjects were 241 persons recruited in the
environment to be rated. We generated a list of as diverse a set of environments as possible, again in or around Vancouver. There was a slight overlap between the sample of environments used in construction of the preliminary scales and the present sample of environments, but for the most part, new environments were selected. This small overlap was not considered a problem, since the present sample of subjects rating the environments was new and even when the same environment was rated, the new subject would presumably not focus on exactly the same aspects of that environment. For example, two subjects recruited in the same urban park are unlikely to have precisely the same experience there. The final set of 241 environments was as diverse as the earlier sample and included as many types of environments as could be thought of (e.g., a cafe, a nearly deserted beach, a bus depot, a discotheque, an urban park, a bowling alley, a student's dormitory room, a farm, a garage, and a suburban street corner). Scale Construction Each subject's data were again ipsatized (equated with respect to mean and variance). Existence of the preliminary scales now allowed a more precise item selection procedure. Items were selected for each scale on the basis of (a) high correlations with the sum
AFFECTIVE QUALITY OF PLACES
317
Table 2 Reliabilities and Intercorrelations Among Eight Unipolar Scales of Affective No.
Quality
Intercorrelations among scales
Range of
nf
Scale
items
a
correlations
2
3
Arousing quality Exciting quality Pleasant quality Relaxing quality Sleepy quality Gloomy quality 7. Unpleasant quality 8. Distressing quality
5 5 5 5 5 S 5 5
.72 .82 .89 .92 .82 .84 .87 .86
.30-.66 .51-.72 .72-78 .7S-.8S .59-.6S .S8-.69 .64-.S2 .64-.7S
.55
-.04 .46
1.
2. 3. 4. S. 6.
4
-.58 -.09 .54
5
6
7
8
-.76 -.48 .04 .58
-.37 -.78 -.60 -.20 .30
.00 -.50 -.79 -.56 -.16 .59
.52 .00 -.57 -.86 -.61 .18 .55
Note. Statistics are based on 241 subjects, after their data were ipsatized (equated with respect to mean and variance). The total score used to calculate item-total correlations did not include the item in question. of the other nine items of that scale and (b) appropriate correlations with the remaining seven scales. The usefulness of the circular ordering of variables for item selection deserves special note. For example, an item for the pleasant quality scale was expected to correlate near zero with arousing quality and sleepy quality variables, both at 90°. Thus, an item correlating positively with arousing quality and negatively with sleepy quality, even if the correlation was low, was not as pure a measure of pleasant quality as was desirable. More generally, each item could be judged on the entire pattern of correlations with all eight scales. In practice, this turned out to be an exceedingly sensitive procedure for selecting items, and the best five items for each scale were easily selected to form the final scales. The number five was chosen to insure adequate reliability while making all scales equal in length and as short as possible. Properties of the Eight Scales Reliability. Properties of the final versions of the eight unipolar scales are given in Table 2. Reliabilites were generally improved over those in Table 1 for the preliminary scales. All were above .80, except the arousing quality scale. Number of dimensions. Principal components analysis of the eight scales (with unities on the diagonal) again supported the hypothesis of precisely two underlying components. Two principal components (with
eigenvalues 3.56 and 3.04) together accounted for 82.4% of the total variance; the next component accounted for a trivial increment in variance (eigenvalue of 0.36). The first two components also accounted for approximately equal proportions of variance; the first unrotated component accounted for 44.4% and the second for 38.0% of the total variance. Circular ordering. The principal compoents were then rotated to an orthogonal Procrustes solution. This final solution is the one shown in Figure 2 and discussed in the Introduction as exemplifying the circular ordering of variables. That is, Figure 2 is an empirical plot of the eight scales located in a two-dimensional space defined by their rotated principal components. Bipolarity of affective quality dimensions. As can be seen in Figure 2, the eight unipolar scales actually fall on four bipolar dimensions. Even though constructed separately and with no overlapping items, each pair of scales was empirically shown to be measuring opposite ends of the same continuum: arousing is opposite sleepy, exciting is opposite gloomy, pleasant is opposite unpleasant, and distressing is opposite relaxing. That these dimensions are bipolar is one criterion for a circular ordering, and this hypothesis predicts that the correlations between such pairs should be -1.00. The obtained correlations, given in Table 2, were - .76, - .78, - .79, and - .86, respectively. With the alpha coefficients used in a correction for attenuation, these figures became - .99 - .94, - .90, and - .96, re-
318
JAMES A. RUSSELL AND GERALDINE PRATT
Table 3 Statistics for Four Bipolar Scales of A/ective Quality No. of items
Scale
1.
Arousing—sleepy
2. Exciting— gloomy 3. Pleasant-unpleasant 4. Distressing-relaxing
10 10 10
10
Correlations among scales
M
SD
a
2
3
5.02 5.80 9.89 -4.45
12.93 13.76 15.20 17.18
.86 .91 .93 .94
.47
-.08 .65
4 .62
-.14 -.65
Note. Statistics are based on a sample of 241. Scales were scored as indicated in Table 4, where items for the scales are also given.
spectively, again clearly showing that the underlying dimensions are bipolar. Correlations among scales. Table 2 gives the full matrix of intercorrelations among the eight scales. As in the case of bipolarity, this matrix closely approximated that expected for a circular ordering of variables and showed improvement over the correlation matrix of Table 1 for the preliminary scales. For example, adjacent scales, which, if exactly 45° apart and measured with error-free data, should have correlated ,707, actually correlated between .30 and .59; orthogonal scales, which should have theoretically correlated .00, actually correlated between —.20 and .18. It is difficult to precisely estimate the degree to which the final scales approximate the model of affective quality outlined earlier. As shown by the principal components analysis illustrated in Figure 2, the eight scales clearly fall in a circular order in a two-dimensional space. Because of measurement error, the scales cannot be expected to be perfectly reliable nor to be exactly 45° apart. Nevertheless, the obtained results appear to be among the closest approximations to a circumplex reported in the personality literature to date (Wiggins, 1979, p. 406). Use of the Scales Bipolar Scales In the practical use of these scales, the eight unipolar scales are formed into four bipolar scales by treating items from one unipolar scale as positively keyed and items from its bipolar opposite scale as negatively keyed items of a single scale. (That is, subjects' re-
sponses to negatively keyed items are multiplied by — 1, and all responses are then summed.) This procedure yields the following four 10-item scales: arousing-sleep, excitinggloomy, pleasant-unpleasant, and distressing relaxing. To examine the properties of these four bipolar scales, we returned to the raw, nonipsatized data (N = 241) used to construct the final scales. Ipsatization is not required for bipolar scales when, as in the present case, response style is eliminated by negatively keying half of the items. Statistics for the four bipolar scales are shown in Table 3. Reliability estimates ranged from .86 to .94 and are highly adequate. Note, however, that these reliability estimates are slightly inflated because they are based on the same sample used for scale construction. Work with the preliminary unipolar scales indicated that these scales closely maintain reliability on cross-validation, but unbiased estimates of the reliability of the four bipolar scales must await a cross-validation. Correlations among the scales, also shown in Table 3, show that these four bipolar scales are interrelated approximately as expected. Arousing-sleepy and pleasant-unpleasant are approximately orthogonal (90°), with a correlation of —.08; similarly, exciting-gloomy and distressing-relaxing are approximately orthogonal, with a correlation of — .14. The other correlations approach the value .707 for dimensions 45° apart or the value — .707 for dimensions 135° apart. In short, the four scales can reliably and validly assess the proposed theoretical structure. Items and instructions for the scales are given in Table 4.
AFFECTIVE QUALITY OF PLACES
These scales measure a person's attribution of affective quality to a particular place by assigning that attribution to a single point in the space depicted in Figure 2. That point can be denned by two values, x and y coordinates. As is customary, we shall define the x and y axes as, respectively, the horizontal (pleasant-unpleasant) axis and the vertical (arousing-sleepy) axis, as shown in Figure 2. Coordinates are then obtained from the following theoretical equations:
x = 1.00 (P-U) + .707 (E-G) - .707 (D-R) (1) and y = 1.00 (A-S) + .707 (E-G) + .707 (D-R), (2) where P-U is the score on the pleasant-unpleasant quality scale, A-S the score on the arousing-sleepy quality scale, E-G the score on the exciting-gloomy scale, and D-R the score on the distressing-relaxing scale. Actually, only two dimensions and hence only two of the four scales are theoretically needed to represent the affective quality space. The ability of the scales given in Table 4 to accomplish this can be empirically tested by examining the ability of two orthogonal scales to predict the remaining two. The following two regression equations use the pleasant-unpleasant (P-U) quality and the arousing-sleepy (A-S) quality scales to predict scores on the remaining two scales:
E-G = .561 A-S + .626 P-U - 3.213 (R = .84) (3) D-R = .756 A-S - .687 P-U - 1.457 (R = .87). (4) Equations 3 and 4 are in raw score form and allow each subject's actual raw score on the criterion (the exciting-gloomy and distressing-relaxing scales) to be predicted from his or her raw scores on the predictors. The multiple correlation (R) is the correlation between subjects' actual scores and those predicted by the equation. These multiple correlations are approaching their upper limit, since they are approaching the reliabilities of the variables involved in the equation.
319
Table 4 Scales of the Affective Places
Quality
Attributed to
Instructions. Below is a list of words that can be used to describe places. We would like you to rate how accurately each word below describes this place. Use the following 1-8 rating scale for your answer. Please be sure that you have given an answer for each word. 1 = extremely inaccurate 5 ! slightly accurate 2 = very inaccurate 6 • quite accurate 3 = quite inaccurate 7 ' very accurate 4 = slightly inaccurate 8 1 extremely accurate Positively keyed items
Negatively keyed items
Arousing quality intense arousing active alive forceful
Sleepy quality inactive drowsy idle lazy slow
Exciting quality exhilarating sensational stimulating exciting interesting
Gloomy quality dreary dull unstimulating monotonous boring
Pleasant quality pleasant nice pleasing pretty beautiful
Unpleasant quality dissatisfying displeasing repulsive unpleasant uncomfortable
Distressing quality frenzied tense hectic panicky rushed
Relaxing quality tranquil serene peaceful restful calm
Note. In actual administration, italicized labels are omitted and items are intermixed in random order. Score on a negatively keyed item is multiplied by —1 when the 10-item bipolar scales are scored.
In practice, this result suggests several possible uses of the scales, (a) When it is important to have as short a questionnaire as possible, the pleasant-unpleasant and arousing-sleepy scales (i.e., a 20-item questionnaire) should suffice to assess the affective quality attributed by an individual to a place. Scores on the other two dimensions may be accurately predicted from these two alone. (Alternatively, the exciting-gloomy and dis-
320
JAMES A. RUSSELL AND GERALDINE PRATT
tressing-relaxing scales could be used—see below.) (b) When a longer (40-item) questionnaire is practical, the reliability of the environmental assessment may be increased by employing all four scales and using Equations 1 and 2. Conscientiousness and accuracy of each rating may then be checked by comparing actual scores on the excitinggloomy and the distressing-relaxing scales with those predicted by Equations 3 and 4. Any large discrepancy would suggest caution and investigation, (c) When the research design calls for retesting on the affective quality scores, pleasant-unpleasant and arousingsleepy may be used for one administration, and exciting-gloomy and distressing-relaxing, for the other. The following equations emphasize this last point by showing how pleasant-unpleasant and arousing-sleepy may be predicted from exciting-gloomy (E-G) quality and distressing-relaxing (D-R) quality: A-S = .533 E-G + .525 D-R + 4.268 (R = .84) (5) P-U = .629 E-G - .509 D-R + 3.975 (R = .86). (6)
Environmental Assessment So far, we have discussed the measurement of each individual's attribution of affective quality to an environment. The results thus bear on the process of environment "perception" (with perception broadly denned). Craik (1971) pointed out that environment perception is related to environment assessment in much the same way that person perception is related to personality assessment. In both personality and environment assessment, the consensus of a group of observers is used as the basis of inferring that an attribute is an attribute of the stimulus (environment or person) rated. To provide preliminary evidence on the amount of consensus among observers rating the affective quality of environments, we assessed the interrater reliability of the pleasant-unpleasant quality and the arousingsleepy quality scales applied to photographs of 20 molar physical environments. These were 5 inch X 7 inch (12.7 cm X 17.8 cm)
prints that have been studied by Ward (1977) and that appeared to represent a reasonable sample of environments. The use of photographic simulations rather than actual environments not only was convenient but insured that subjects were responding to exactly the same stimulus. Moreover, available evidence indicates that simulations, especially if they include visual input, provide a surprisingly good approximation to actual environments. Subjects were 609 visitors to an open house at the University of British Columbia. Visitors were of both sexes and of all ages. Those visitors who appeared to be at least 13 years old were asked to volunteer to serve as a subject in a study. Each volunteer randomly selected 1 of the 20 photographs and described it on the scales of pleasant-unpleasant and arousing-sleepy qualities (Table 4). Each photograph was rated by at least 16 subjects; the maximum number of subjects was 39 and the mean was 30.4. For each photograph, the sample of subjects was randomly divided in half to calculate across the 20 photographs a split-half estimate of the interrater reliability of each scale. The obtained figures, corrected by the SpearmanBrown formula, were .97 for both scales, indicating that there does indeed exist a consensus among observers on the affective quality of a place. Discussion This study was not an attempt to discover relationships among affective variables. Rather, on the basis of previous empirical work, a theoretical structure was hypothesized in advance. This study was carried out to evaluate the usefulness of that structure in terms of (a) the degree to which further empirical data could be represented by that theoretical structure and (b) whether valid assessment techniques could be developed from it. By these criteria, the structure appears valuable. Simple, reliable, and valid scales were developed to assess the affective quality attributed to places, and the relationship among these scales corresponded within the margin of reliability of psychological data to the proposed theoretical structure. To-
AFFECTIVE QUALITY OF PLACES
gether with previous evidence (e.g., Russell et ah, Note 1), the present evidence thus indicates that the proposed concepts and scales adequately describe the affective quality attributed to places. These concepts and scales should prove valuable to environmental (and perhaps social) psychologists in a number of ways. They may be used as dependent variables in an attempt to understand what objectively specifiable properties of environments (or social situations) produce a given affective reaction. This concern is exemplified by Berlyne's (1971, 1974) work on aesthetics, in which he investigated how analogues of pleasure and arousal reactions varied as a function of various stimulus parameters. These scales could also be used as predictor variables in an attempt to explain behavioral reactions to environments or situations. This concern is exemplified by work on the behavioral effects of stressors (e.g., Stokols, 1972); by Patterson's (1978) work on behavioral consequences of arousal; by Freedman's (1975) notion that pleasantness interacts with crowding to influence behavior; or, more generally, by Mehrabian and Russell's (1974) proposal that the affective reaction to places influences diverse behaviors, including social interaction, drug consumption, and work performance. On a conceptual level, it may be possible to help integrate the empirical findings of these various investigations by viewing their findings within the broader framework offered by the proposed model. The proposed conceptualization may also prove to be valuable in designing and evaluating research. The results with distressing quality, for example, offer suggestions on the notion of psychological stress, which has been used by some investigators to mean any response to a negative (unpleasant) situation but by others as synonymous with arousing quality. Defined so broadly it is no wonder that diverse results are obtained in studies of environmental stressors. The proposed conceptualization suggests that stress might better be defined as occurring only when the situation is both unpleasant and arousing to the subjects involved. The scales of Table 4 should provide a means of actually assessing
321
the subject's perception of the "stressful" situation. Moreover, the perceived affective quality of the "nonstressful" control condition is equally important in determining the results of experiments on stress. Suppose that in one study, the stressor is highly arousing and only slightly unpleasant and the control condition is unarousing but also slightly unpleasant (e.g., the control group receives a slightly boring task). In this case, the difference between the experimental and control conditions would actually be a difference in arousal with no difference in pleasantness. In a second experiment on stress, it is conceivable that the stressor would be highly unpleasant and only moderately arousing, and the control condition would be pleasant but also moderately arousing (e.g., an interesting task). In this second case, the difference between experimental and control conditions would be mainly a difference in pleasantness with no difference in arousal. There is every reason to expect that these two experiments on stress should produce different results. The same reasoning suggests that in this way improvements can be obtained in research on the effects not only of stress but of arousal, anger, aversive events, discomfort, mood, positive-negative affective states, or any of the other affective variables increasingly studied in social psychological research. In short, the proposed description of affective quality should offer a means of replacing broad and vaguely defined two-valued categories (such as stressful vs. nonstressful) with a conceptualization capable of more precise specification. Reference Note 1. Russell, J. A., Ward, L. M., & Pratt, G. A factor analytic study of the affective quality attributed to environments. Unpublished manuscript, 1978. (Available from J. A. Russell, University of British Columbia, 2075 Wesbrook Mall, Vancouver, British Columbia, Canada V6T 1W5.)
References Benjamin, L. S. Sructural analysis of social behavior. Psychological Review, 1974, 81, 392-425. Bentler, P. M. Semantic space is (approximately) bipolar. Journal of Psychology, 1969, 71, 33-40. Bentler, P. M. An analysis of responses to adjectives:
322
JAMES A. RUSSELL AND GERALDINE PRATT
A reply to Samelson. Psychological Bulletin, 1973, 80, 133-134. Berlyne, D. E. Aesthetics and psychobiology. New York: Appleton-Century-Crofts, 1971. Berlyne, D. E. (Ed.). Studies in the new experimental aesthetics: Steps toward an objective psychology of aesthetic appreciation. New York: Wiley, 1974. Craik, K. H. The assessment of places. In P. McReynolds (Ed.), Advances in psychological assessment (Vol. 2). Cupertino, Calif.: Science & Behavior Books, 1971. Foa, U. G., & Foa, E. B. Societal structures of the mind. Springfield, 111.: Charles C Thomas, 1974. Freedman, J. L. Crowding and behavior. San Francisco: Freeman, 1975. Glass, D. C., & Singer, J. E. Urban stress: Experiments on noise and social stressors. New York: Academic Press, 1972. Guttman, L. A new approach to factor analysis: The radex. In P. R. Lazarsfeld (Ed.), Mathematical thinking in the social sciences. New York: Free Press of Glencoe, 1954. Ittelson, W. H. Environment perception and contemporary perceptual theory. In W. H. Ittelson (Ed.), Environment and cognition. New York: Seminar, 1973. Leary, T. Interpersonal diagnosis of personality. New York: Ronald Press, 1957. Lorr, M., & McNair, D. M. An interpersonal behavior circle. Journal of Abnormal and Social Psychology, 1963, 67, 68-75. Meddis, R. Bipolar factors in mood adjective checklists. British Journal of Social and Clinical Psychology, 1972, 11, 178-184. Mehrabian, A., & Russell, J. A. An approach to environmental psychology. Cambridge, Mass.: MIT Press, 1974. Milgram, S. The experience of living in cities. Science, 1970, 167, 1461-1468.
Osgood, C. E. On the whys and wherefores of E, P, and A. Journal of Personality and Social Psychology, 1969, 12, 194-199. Patterson, M. L. Arousal change and cognitive labeling: Pursuing the mediators of intimacy exchange. Environmental Psychology and Nonverbal Behavior, 1978, 3, 17-22. Russell, J. A., & Mehrabian, A. Evidence for a three-factor theory of emotions. Journal of Research in Personality, 1977, ;;, 273-294. Russell, J. A. Evidence of convergent validity on the dimensions of affect. Journal of Personality and Social Psychology, 1978, 36, 1152-1168. Russell, J. A. Affective space is bipolar. Journal of Personality and Social Psychology, 1979, 37, 345356. Schaefer, E. S. A circumplex model for maternal behavior. Journal of Abnormal and Social Psychology, 1959, 59, 226-235. Schaefer, E. S., & Plutchik, R. Interrelationships of emotions, traits, and diagnostic constructs. Psychological Reports, 1966, 18, 399-419. Schlosberg, H. Three dimensions of emotion. Psychological Review, 1954, 61, 81-88. Stokols, D. On the distinction between density and crowding. Psychological Review, 1972, 79, 275-277. Ward, L. M. Multidimensional scaling of the molar physical environment. Multivariate Behavioral Research, 1977, 12, 23-42. Wiggins, J. S. Personality and prediction: Principles of personality assessment. Reading, Mass.: Addison-Wesley, 1973. Wiggins, J. S. A psychological taxonomy of traitdescriptive terms: The interpersonal domain. Journal of Personality and Social Psychology, 1979, 37t 395-412. Received February 21, 1979 •
