Relating momentary affect to the five factor model of personality: A Japanese case 1

Japanese Psychological Research 2003, Volume 45, No. 2, 80 93 Munksgaard ORIGINAL ARTICLE Relating momentary affect to the five factor model of personality: A Japanese case 1 MICHELLE S. M. YIK 2 Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong JAMES A. RUSSELL Boston College, Chestnut Hill, Massachusetts, USA NAOTO SUZUKI Department of Psychology, Faculty of Letters, Doshisha University, Karasuma Imadegawa, Kamigyo-ku, Kyoto 602-8580, Japan Abstract: This article shows that fundamental aspects of the structure of momentary affect are similar in Japanese and Canadian societies. We developed questionnaire scales in Japanese in four different formats for assessing self-reported momentary affect. Scales can be scored for dimensions defined by Feldman Barrett and Russell (1998), Thayer (1996), Larsen and Diener (1992), and Watson and Tellegen (1985). We administered these newly developed affect scales and NEO five-factor inventory (Costa & McCrae, 1992) to a sample of 450 Japanese respondents. The affect scales were found to be psychometrically sound and to be interrelated, as found with English-speaking Canadians. Dimensions could be integrated into a two-dimensional affective space. Personality correlated with momentary affect, though not in the same pattern as found in Canada. Key words: Strucutre of Affect, Five Factor Model of Personality, Japanese. For theoretical and practical reasons, psychologists are increasingly turning to the study of affect. Affect has been found to be an important factor in research laboratory, clinic, advertising, and workplace. There are three interrelated purposes in the present study. First, we test the generalizability to Japanese of the two-dimensional affective model developed in English. Second, we develop ready-to-use tools to assess momentary affect in Japanese. Finally, we conduct a cross-language comparison on the connections between momentary affect and personality. A circumplex model of affect In the past decade, various dimensional models have been proposed to describe the covariations of self-reported affective feelings in English. Major models include Russell s (1980) circumplex, Thayer s (1996) energetic and tense arousal, Larsen and Diener s (1992) eight combinations of pleasantness and activation, and Watson and Tellegen s (1985) Positive Affect and Negative Affect. These models were shown to fit comfortably within a two-dimensional affective space, characterized by two bipolar axes 1 Data reported in this article are part of Michelle Yik s doctoral dissertation. We thank Yuka Mukai and Yuriko Takahasi for their help in translation. 2 Correspondence concerning this article should be sent to: Michelle Yik, Hong Kong University of Science and Technology, Division of Social Science, Clear Water Bay, Kowloon, Hong Kong (Email: myik@ust.hk). 2003 Japanese Psychological Association. Published by Blackwell Publishing Ltd.

Momentary affect in the Japanese 81 Figure 1. A circumplex model of affect. Fourteen unipolar affect constructs empirically placed in an integrated twodimensional space via CIRCUM (Browne, 1992). Results are obtained from a study of 535 Englishspeaking Canadians. Adopted from Russell, Yik, and Steiger (in press). of Pleasant versus Unpleasant, and Activated versus Deactivated (Feldman Barrett & Russell, 1998; Yik, Russell, & Feldman Barrett, 1999). Figure 1 shows an empirical example of that integrated space in English (Russell, Yik, & Steiger, in press). On the right hand side are the pleasant states; on the left hand side are the unpleasant states. On the upper half are the activated states; on the lower half are the deactivated states. Within this two-dimensional space, any specific affective state is composed of different levels of Pleasantness and Activation. This space is a circumplex in which affective dimensions fall in a circular ordering along the perimeter. The circumplical nature of affective states has received strong empiric support (Remington, Fabrigar, & Visser, 2000). We do not, however, assume that the structure of Figure 1 captures all of affect. Rather, we propose it as a means of representing affect at a general level. One question immediately arises: Can the integrated space be generalized to other languages, such as Japanese? Initial support for the integrated structure was obtained from previous studies in Japanese. Ogawa, Takehara, Monchi, Fukui, and Suzuki (1999) found that the two-dimensional space was useful to describe the ratings of facial expressions under ambiguous perceptual conditions (see also Ogawa & Suzuki, 2000). Affect scales were also developed to measure some vectors of the proposed integrated space (Ogawa, Monchi, Kikuya, & Suzuki, 2000). In the present study, scales defining the affective dimensions in Figure 1 were translated into Japanese. The resulting tools are valuable for pursuing each of the original models in Japanese. Together, they also allowed us to examine whether they can be integrated in the way indicated as they are in Figure 1. Integration relied on confirmatory factor analysis, a powerful tool which estimates relations among variables relatively free of random and systematic errors inherent in measurement. To take full advantage of confirmatory factor analysis requires that each dimension be measured in several different ways. We therefore developed all scales in at least three different response formats. Predicting affect from personality Our study also explored the link between momentary affect (state) and personality (trait). Momentary affect obviously can be predicted

82 M. S. M. Yik, J. A. Russell and N. Suzuki by immediate context, but it is also obvious that some people typically feel anxious, others typically feel happy, and still others typically feel relaxed. In other words, one s momentary affect can be predicted from one s personality (e.g., Diener, 1984; Larsen & Ketelaar, 1991; McCrae & Costa, 1991). In the present study, we use the Japanese translation of NEO Five Factor Inventory (NEO FFI; Costa & McCrae, 1992) to measure personality. Much work has linked Extraversion (E) and Neuroticism (N) to Watson and Tellegen s (1985) Positive Affect and Negative Affect, respectively. The robustness of the findings led Tellegen (1985; Watson & Clark, 1984) to suggest that E and N should be relabeled as Positive Emotionality and Negative Emotionality. Fewer studies have been conducted to examine the predictive utility of Agreeableness, Conscientiousness, and Openness to Experience on affect (Costa & McCrae, 1984; McCrae & Costa, 1991; Watson & Clark, 1992). Positive relations were reported between Openness to Experience and positive affective states. Both Agreeableness and Conscientiousness were found to correlate positively with positive affective states and negatively with negative affective states. The circumplex model of Figure 1 provides a simple but powerful way to summarize relations between the affect variables and any outside variable, such as a personality variable. The principle is that any outside variable which correlates reliably with one affect variable will correlate with the remaining affect variables and the pattern of correlations will form a sine wave. That is, the magnitude of the correlations will rise and fall in a sine pattern. Thus, it is the appearance of the sine wave, rather than the statistical significance of individual correlations, which speaks directly to the integrity and utility of the circumplex model of affect. Overview of the present study Data from 450 Japanese respondents were used to examine three issues: (a) The psychometric properties of the translated affect scales; (b) the structure among those scales, specifically the circumplex ordering; and (c) the relation of momentary affect to personality. Method Participants Participants were 450 undergraduates (228 men, 222 women) from Doshisha University. Their mean age was 19.69 (SD = 1.15). Participation was voluntary. Test administration took place during class time. Procedure Participants first completed an affect questionnaire under the title Remembered Moments Questionnaire and then a personality questionnaire under the title NEO FFI. All questionnaires were in Japanese. Affect questionnaire Translation. All instructions and scales were translated into Japanese by two bilinguals. A back-translation procedure was adopted. First, one bilingual translated the English version of the affect questionnaire (Yik, Russell, & Feldman Barrett, 1999) into Japanese. Second, another bilingual, who was blind to the English original, translated the Japanese version back into English. Translations were revised until satisfactory before we used them in the data collection. 3 Instructions. The front page of the battery provided general instructions under the title Remembered Moments Questionnaire. There were six versions of the questionnaire, each with a different anchoring time. The six anchoring times were before breakfast, after breakfast, before lunch, after lunch, before dinner, and after dinner. Participants were randomly assigned to one of the six instructions. For instance, the instructions for one version were as follows: 3 The backtranslation procedure is not without problems. The selection of equivalents is a matter of judgment. Different translators and different researchers routinely produce slightly different translations. In our case, we made our judgment call. We used data analysis as a means of rejecting items that failed to behave properly.

Momentary affect in the Japanese 83 we need to ask you to remember a particular moment. Please think back to yesterday. Specifically, recall the time just before breakfast. (If you didn t have breakfast yesterday, simply recall that approximate time of day.) It is important that you remember a specific moment accurately. So, please search your memory and try to recall where you were, what you were doing at that time, who you were with, and what you were thinking. Now select a particular moment that is especially clear in your memory. (If you really have no recollection of the time just before breakfast, please search your memory for the closest time that you do recall accurately.) In the other five versions, italicized words were replaced. The instructions then emphasized that all subsequent questionnaires were to be answered with respect to that selected moment of the day before. Formats. Participants then received a battery of four questionnaires, each in a different format, in the following order: (a) Semantic differential scales (SEM); (b) adjective format (ADJ), which was an adjective list accompanied by a five-point Likert scale ranging from 1 not at all to 5 extremely ; (c) Agree-Disagree format (AGREE), a list of statements with which participants were asked to indicate their degree of agreement, ranging from 1 strongly disagree to 5 strongly agree ; and (d) Describes Me format (DESCRIBE), a list of statements, for each of which participants were asked to indicate how well it described their feelings, ranging from 1 not at all to 4 very well. The SEM format consisted of bipolar measures of Pleasure and Arousal translated directly from Mehrabian and Russell (1974). The remaining three questionnaires were unipolar in format and each questionnaire included translated items from (a) Feldman Barrett and Russell s (1998) Current Mood Questionnaire (CMQ) assessing Pleasant, Unpleasant, Activated, and Deactivated affect; (b) Larsen and Diener s (1992) Activated Unpleasant, Unactivated Unpleasant, Activated Pleasant, and Unactivated Pleasant affect; (c) Thayer s (1996) Energy, Tiredness, Tension, and Calmness; and (d) Watson, Clark and Tellegen s (1988) Positive Affect and Negative Affect. Therefore, scores on these various scales could be derived from our sample. The Japanese version of all affect measures is available from the first author upon request. Personality measure Our measure of personality was a 60-item Japanese translation of NEO Five Factor Inventory (NEO FFI; Costa & McCrae, 1992) developed by Shimonaka, Nakazato, Gondo, and Takayama (1999). NEO FFI is part of their Japanese translation of the 240-item NEO Personality Inventory Revised (NEO PI-R; Costa & McCrae, 1992) and is intended to offer an abbreviated measure for revised NEO personality inventroy (NEO PI-R). Responses are made on a 5-point rating scale ranging from Strongly Disagree through to Neutral to Strongly Agree. Each of the five NEO FFI scales consisted of 12 items. Cronbach s alphas for the five personality scales ranged from 0.63 to 0.85, indicating that they are internally consistent. The five scales showed the expected small, but reliable, correlations with each other. These psychometric properties resemble those found in the administration of NEO FFI in a similar Japanese sample (Shimonaka, et al., 1999) and with the original English version (Costa & McCrae, 1995). Results and Discussion The results are presented in four sections. First, we examine psychometric properties of a priori scales. Second, we revise the scales to more closely approximate the desired structure, and report psychometric properties of the revised scales. Third, we examine the integrated structure among the affect constructs originating from different structural models. Fourth, we examine the relation between affect (based on the revised scales) and personality.

84 M. S. M. Yik, J. A. Russell and N. Suzuki Table 1. Indices of fit for a priori and revised measurement models (N = 450) Model χ 2 df RMSEA AGFI CFI CMQ a priori Model with Correlated Constructs 194.62 30 0.11 0.83 0.96 Model with Correlations between Constructs fixed to zero 829.02 36 0.20 0.56 0.80 CMQ Revised Model with Correlated Constructs 173.40 30 0.10 0.85 0.95 Model with Correlations between Constructs fixed to zero 709.73 36 0.19 0.61 0.79 Thayer s Unipolar Constructs a priori Model with Correlated Constructs 185.93 30 0.11 0.84 0.96 Model with Correlations between Constructs fixed to zero 688.10 36 0.19 0.60 0.85 Larsen & Diener s Unipolar Constructs a priori Model with Correlated Constructs 170.45 30 0.10 0.85 0.97 Model with Correlations between Constructs fixed to zero 797.19 36 0.21 0.54 0.85 Watson & Tellegen s Unipolar Constructs a priori Model with Correlated Constructs 161.13 5 0.25 0.58 0.93 Model with Correlations between Constructs fixed to zero 161.38 6 0.23 0.65 0.93 Constructs fixed to zero. CMQ, Current Mood Questionnaire; RMSEA, root mean square error of approximation; AGFI, adjusted goodness of fit; CFI, comparative fit index. Individual measurement models In this section, we examine the ability of the various a priori affect scales to assess the original four structures from which the scales were developed. To examine how well each of the four structural models fit the Japanese data, we used a confirmatory factor analysis, with each latent construct indicated by three scales with different response formats. For all factor models, we estimated: (a) factor loading between each manifest variable and its intended latent construct; (b) error term associated with each manifest variable; (c) correlation between error terms with the same response format; and (d) correlations between latent constructs. Table 1 gives indices of fit for all the models. Models hypothesized by Feldman Barrett and Russell (1998), Thayer (1996), and Larsen and Diener (1992) all fit the data well, significantly better than did a comparison model in which the correlations among latent constructs were fixed to 0.00. In contrast, the hypothesized model for Watson and Tellegen (1985) fit the data poorly. Yik et al. (1999) found that bipolar versions of Watson and Tellegen s dimensions (Positive Affect and Negative Affect) produced a reasonable measurement model. To replicate the findings of Yik et al., Larsen and Diener s Unactivated Unpleasant was used as the bipolar opposite of Positive Affect, Unactivated Pleasant the opposite of Negative Affect. To assess the bipolar version of Watson and Tellegen s model, we specified a confirmatory factor analysis with four latent constructs, each indicated by three scales with different response formats. The hypothesized model fit the data modestly: χ 2 (30, N = 450) = 318.55, root mean square error of approximation (RMSEA) = 0.15, adjusted goodness of fit (AGFI) = 0.73, comparative fix index (CFI) = 0.94. This model also fit the data significantly better than the comparison model: χ 2 (6, N =

Momentary affect in the Japanese 85 Table 2. Measurement models for a priori (and revised) Current Mood Questionnaire (CMQ) scales: Confirmatory factor analyses (N = 450) Construct Format Pleasant Unpleasant Activated Deactivated M SD Standardized Factor Loading Pleasant ADJ 0.90* 2.76 1.16 (.87*) (2.83) (1.23) Pleasant AGREE 0.97* 3.21 1.12 (.96*) (3.27) (1.13) Pleasant DESCRIBE 0.75* 2.41 0.82 (.82*) (2.58) (1.01) Unpleasant ADJ 0.93* (.90*) 1.78 (1.85) 0.94 (1.05) Unpleasant AGREE 0.90* (.89*) 1.93 (1.93) 1.08 (1.08) Unpleasant DESCRIBE 0.90* (.88*) 1.85 (1.85) 0.89 (.89) Activated ADJ 0.78* (.58*) 2.18 (2.17) 0.97 (1.03) Activated AGREE 0.86* (.58*) 2.36 (2.55) 0.89 (1.09) Activated DESCRIBE 0.61* (.41*) 1.78 (1.67) 0.63 (.87) Deactivated ADJ 0.71* (.65*) 2.54 (2.74) 0.91 (1.01) Deactivated AGREE 0.62* (.56*) 2.64 (2.64) 0.87 (.87) Deactivated DESCRIBE 0.85* (.91*) Interfactor Correlation Pleasant ( ) Unpleasant.78* (.81*) ( ) Activated.31* 0.63* (.18*) (.08) ( ) Deactivated 0.00.10.49* (.11*) (.06) (.72*) ( ) 2.31 (2.23) Figures without parentheses are estimates for a priori CMQ scales; figure in parentheses are estimates for revised CMQ scales. Possible mean scores range from 1 to 5 for Adjective and Agree-Disagree formats; 1 4 for Describes Me format. p 0.01. 0.66 (.77) 450) = 514.89, p < 0.001, and RMSEA changed from 0.15 to 0.22. To sum up, the structures proposed by each of the original authors based on English data was supported in the present Japanese data, with some difficulties encountered for Watson and Tellegen s (1985) variables. Individual constructs were adequately measured by the three scales with different response formats. Constructs were related to each other approximately as predicted. Since the Feldman Barrett and Russell (1998) model forms the core of the proposed integrated space, the model estimates are given in Table 2. Parameter estimates for other structural models

86 M. S. M. Yik, J. A. Russell and N. Suzuki are available from the first author upon request. Revising the affect scales Because the scales defining the 14 affect constructs are borrowed directly from English by translation, it is possible that these a priori scales were not adequate indicators of the intended underlying latent constructs and might lack adequate internal consistency. Such psychometric problems could obscure the structural appearance of the models tested and may worsen the fit indices and thus the measurement models. In order to define the latent constructs and thus the structural models tested by culturally appropriate scales, efforts are therefore directed to revising the scales on the basis of measurement models and reliability analyses. Revising the Current Mood Questionnaire scales. Pleasant, Unpleasant, Activated, and Deactivated these four affect constructs are the cornerstones of the two-dimensional space proposed in the present study. Thus, it is of paramount importance to ensure that the pattern of correlations among these four aligns with that expected. In order to ensure the proper placements of the four constructs within the two-dimensional space, the Pleasant and Unpleasant scales must be relatively independent of the Activation dimension (neither activated nor deactivated) and that Activated and Deactivated scales must be relatively independent of the Pleasantness dimension (neither pleasant nor unpleasant). We thus revised these 12 scales (3 scales 4 constructs) with two interrelated purposes in mind: (a) the negative correlations between the hypothesized bipolar opposites (Pleasant vs. Unpleasant, Activated vs. Deactivated) should be maximized; and (b) the correlations between non-bipolar pairs (such as Pleasant vs. Activated) should be minimized. Any revision procedure can be accused of capitalizing on chance. We therefore took steps to minimize this possibility: (a) No items were allowed to switch from one scale to another; and (b) items could only be dropped (but not added) in the revision procedure. With these criteria in mind, we found that revisions were needed. Of 12 scales (4 constructs 3 response formats), 10 were revised. As shown in Table 1, the revised CMQ scales fit the hypothesized model better than had the a priori scales. Further, the revised CMQ model fit the data significantly better than did a comparison model with correlations among latent constructs fixed to 0.00: χ 2 (6, N = 450) = 536.33, p < 0.001, and RMSEA changed from 0.10 to 0.19. Parameter estimates are given in parentheses in Table 2 and the revised scales were used in subsequent analyses. Revising other affect scales. To maintain the similarity between the Japanese and English versions of the scales developed by other authors, we used a more conservative procedure in the present revision. Reliability estimates for scales defining Thayer s (1996), Larsen and Diener s (1992), and Watson and Tellegen s (1985) constructs were examined. The purpose was to make sure that the scales were internally consistent (with a minimum alpha of 0.60) and that they were a reasonable indicator of the intended constructs (with a minimum factor loading of 0.70). All but one of Thayer s scale passed the reliability criterion; one item was dropped from that scale which was used in subsequent analyses. All scales passed the 0.70 loading criterion. The full two-dimensional affective space A correlation matrix including all the affect scales examined here showed substantial correlations across different structural models. Our hypothesis was that all structures are mappable onto one common integrated space. In the following, we adopted two ways to test this convergence across structures. Structural equation models. One way to demonstrate the convergence of constructs of different origins was to use Pleasant versus Unpleasant and Activated versus Deactivated axes as exogenous variables to predict all other affect constructs. By treating all other constructs as endogenous variables, we could

Momentary affect in the Japanese 87 test the hypothesis that the two axes explain most of the reliable variance in other affect constructs. First, we specified a confirmatory factor model (what we call Model 1) with two latent constructs, corresponding to the bipolar axes of Pleasant versus Unpleasant and Activated versus Deactivated affect. Each latent construct was indicated by the bipolar versions of its three scales with different response formats. The semantic differential scale of Pleasure was specified to load on the Pleasant versus Unpleasant construct; the semantic differential scale of Arousal was to load on the Activated versus Deactivated construct. The latent correlation between the two axes was fixed to 0.00. Model 1 fit the data moderately well: χ 2 (16, N = 450) = 111.86, RMSEA = 0.12, AGFI = 0.86, CFI = 0.97. Model estimates were used in subsequent analyses. In the following structural equation models, parameter estimates on the exogenous side were adopted from Model 1 in the preceding paragraph. In each analysis, we estimated (a) loading between a manifest variable and the endogenous construct; (b) regression weights of the endogenous construct on the exogenous constructs; and (c) percentage of variance explained by the exogenous constructs for each endogenous construct. We conducted a separate analysis for each of the six bipolar constructs (2 from Thayer, 2 from Larsen and Diener, and 2 from Watson and Tellegen). Results are summarized in Table 3. All affect constructs were substantially explained by the two axes. The mean variance explained was 92% for the pleasant activated affects and 91% for the unpleasant activated affects. The pattern of relations between the exogenous variables and the endogenous variables was approximately as expected in Figure 1. Consistent with results obtained in English, the four structures can be comfortably integrated into a two-dimensional space. CIRCUM. Another way to demonstrate the convergence of the constructs across different models was to portray the full representation of all constructs simultaneously within a two-dimensional space. To do so, we used CIRCUM, a structural equation modeling program developed by Browne (1992) to Table 3. Affect Constructs Explained by the Revised Pleasant versus Unpleasant and Activated versus Deactivated Axes (N = 450) Regression Weight Indices of Fit Construct Pleasant vs. Unpleasant Activated vs. Variance Deactivated explained (SE) χ 2 RMSEA AGFI CFI Pleasant Activated vs. Unpleasant Deactivated affect Energy vs. Tiredness 0.64 0.69 89 (1.6) 299.84 0.10 0.87 0.95 Activated Pleasant vs. 0.72 0.65 94 (1.3) 241.37 0.09 0.90 0.96 Unactivated Unpleasant High vs. Low Positive Affect 0.61 0.74 93 (1.5) 296.65 0.10 0.87 0.95 Unpleasant Activated vs. Pleasant Deactivated affect Tension vs. Calmness.74 0.58 88 (1.7) 326.27 0.11 0.86 0.95 Activated Unpleasant vs..87 0.41 92 (1.2) 297.33 0.10 0.87 0.96 Unactivated Pleasant High vs. Low Negative Affect.87 0.40 92 (1.2) 283.44 0.09 0.88 0.96 All regression coefficients are significant at.001 level, which is equivalent to an overall alpha level of less than.01 (by Bonferroni procedure, 12 regression weights 0.001 = 0.012). Thayer (1996); Larsen and Diener (1992); Watson and Tellegen (1985). RMSEA, root mean square error of approximation; AGFI, adjusted goodness of fit; CFI, comparative fit index.

88 M. S. M. Yik, J. A. Russell and N. Suzuki Figure 2. Japanese scales of momentary affect. A circumplex representation of 14 unipolar affect constructs via CIRCUM (Browne, 1992). Communality was left free to vary. Figures given are estimates of polar angles with the 95% confidence intervals in parentheses. examine how well our data conformed to a circumplex structure. This program provides fit indices and angular position for each affect variable. This analysis uses the unipolar affect scales. (The semantic differential scales were not used in this analysis.) First, a score was created for each of the 14 unipolar constructs by summing the z-scores of its three scales with different response formats. A 14 14 correlation matrix was then computed with the resulting sums and was submitted to the maximum likelihood estimation using CIRCUM. Pleasant was designated as the reference variable (its location was fixed at 0 ). The locations of other variables were then estimated relative to Pleasant. The communality estimates of all variables were left free to vary. No constraints were put on the minimum common score correlation. The analysis converged on a solution in 12 iterations. Three free parameters were specified in the correlation function equation; additional free parameters did not improve the model fit. The final model had a total of 44 free parameters and 61 degrees of freedom. The data fit the model only moderately well: χ 2 (61, N = 450) = 355.45, RMSEA = 0.10. The results are shown in Figure 2. The four cornerstone variables (Pleasant, Unpleasant, Activated, and Deactivated) were located close to the predicted values: With Pleasant fixed at 0, Activated was 76 away, Unpleasant was 167 away, and Deactivated was 262 away. Hypothesized bipolar opposites were located close to the predicted values: Pleasant was 167 from its bipolar opposite, Unpleasant. Activated was 186 from its bipolar opposite, Deactivated. Constructs developed by various authors fell remarkably close to what we see in Figure 1.

Momentary affect in the Japanese 89 Summary. We set out to test the hypothesis of convergence of affect constructs across different structural models by structural equation models and the CIRCUM analyses. Results from both analyses showed that a twodimensional space defined by Pleasant versus Unpleasant and Activated versus Deactivated axes was able to integrate affect constructs originating from different authors. Further, variables fell at various angles within the space. The results are consistent with a circumplex, which predicts variables to fall at any place along a circle. Relating affect to personality In this section, we examine the connections between affect and personality. First, we examine the ability of personality in predicting momentary affect in a series of structural equation models. Second, we place each personality dimension into the integrated affective space. Predictive utility of personality. To investigate the ability of personality to predict affect, we treated personality as exogenous and affect as endogenous. All exogenous variables were manifest variables. Each of the eight bipolar affect variables served, in turn, as an endogenous variable. In all, we computed 16 structural equation models. For all structural equation models, we estimated: (a) loading between a manifest variable and the endogenous construct; (b) error term associated with each manifest variable; (c) regression weights of the exogenous variables on the endogenous variables; and (d) a percentage of variance explained by the exogenous variables for each endogenous construct. The question was, for each bipolar affect variable, which combination of personality dimensions was the best predictor. In Table 4, we compared two: the E-and-N Model (E and N) versus the FFM (FFM; N, E, O, A, and C). The E-and-N model fit the data well. The mean RMSEA was 0.03. The variance explained ranged from 2.7% to 6.7%, with a mean of 5.0%. The FFM fit the data even better. The mean RMSEA was 0.02. The variance explained ranged from 4.1% to 8.0%, with a mean of 6.5%. Because the E-and-N model was nested with the FFM, we report the χ 2 difference between the two. In all eight cases, the χ 2 change statistic favored the FFM. Only one case showed a significant difference indicating a reliable improvement in model fit by adding O, A, and C to E and N. Consistent with findings in English, the five personality factors yielded higher predictive ability than did the E and N alone. Structural convergence of affect and personality. The circumplex provides a powerful prediction about the pattern of correlations between the set of affect variables and any outside variable. The correlations between any particular personality dimension and all 14 affect constructs should fit a sine function. A sine wave can then be used to evaluate the integrity of the twodimensional space and to locate the personality dimensions within that space. We therefore fit a sine function to each set of correlation relating a personality dimension with all 14 affect constructs. We first computed a score for each of the 14 unipolar affect constructs by summing the z- scores of its three constituent scales. Second, we computed correlations between each FFM dimension and the affect constructs. In Figure 3, the affect dimensions are represented on the abscissa by their angles within the circumplex derived from the CIRCUM analyses described in the preceding section. A sine function fit the data well for all personality dimensions. Browne s (1992) CIRCUM procedure provides a maximum likelihood estimate of where within the two-dimensional affective space each personality dimension falls. For each personality dimension, three figures are given. The angle estimates the location within the circumplex for the personality variable. The zeta estimates the correlation between the personality dimension and the affect vector at the angle specified. The VAF examines the model fit. Results are shown in Table 5. Comparable values from an English sample are also shown. Personality and momentary affect were linked in different ways between Japanese and English.

90 M. S. M. Yik, J. A. Russell and N. Suzuki Table 4. Predicting affect from personality: A comparison of Japanese with English (N = 450) Regression weight N E O A C Variance explained by E-and-N s model (SE) Variance Explained by FFM (SE) Two Axes Pleasant vs. Unpleasant.15 0.01 0.02 0.17* 0.04 4.9 (2.0) 7.5 (2.5).32* 0.04 0.06 0.06 0.05 13.0 (2.8) 13.9 (2.8) Activated vs. Deactivated 0.01 0.20* 0.05.03 0.08 4.4 (2.2) 5.2 (2.3) 0.08 0.18* 0.05.03 0.06 3.0 (1.5) 3.6 (1.6) Pleasant Activated vs. Unpleasant Deactivated affects Energy vs. Tiredness.12 0.14 0.05 0.05 0.11 6.5 (2.3) 8.0 (2.5).17* 0.04 0.06.01 0.07 4.0 (1.7) 4.7 (1.8) Activated Pleasant vs. Unactivated Unpleasant.13 0.14 0.07 0.05 0.08 6.7 (2.3) 8.0 (2.5).20* 0.07 0.10 0.01 0.07 6.5 (2.1) 8.0 (2.3) High vs. Low Positive Affect.13 0.13 0.08 0.04 0.08 5.6 (2.2) 6.9 (2.4).21 0.06 0.13 0.01 0.12 7.1 (2.2) 9.8 (2.5) Unpleasant Activated vs. Pleasant Deactivated affect Tension vs. Calmness 0.15 0.08.03.12.03 2.7 (1.6) 4.1 (1.9) 0.34* 0.08.02.04 0.04 10.1 (2.5) 10.5 (2.6) Activated Unpleasant vs. Unactivated Pleasant 0.19* 0.04.01.14 0.01 4.5 (2.0) 6.1 (2.2) 0.38* 0.07.05.04 0.00 13.4 (2.8) 13.8 (2.8) High vs. Low Negative Affect 0.19* 0.04.03.14.02 4.5 (2.0) 6.3 (2.3) 0.38* 0.07.04.06.01 13.8 (2.8) 14.3 (2.8) Figures in italic are results in an English sample of N = 535 (Yik, 1998). Figures in parentheses are the standard errors. N, Neuroticism; E, Extraversion; O, Openness to Experience; A, Agreeableness; C, Conscientiousness. Significance level was set at.001 in order to achieve an overall alpha of less than.05 (by Bonferroni procedure, 40 regression weights 0.001 = 0.04). Feldman Barrett and Russell (1998); Thayer (1996); Larsen and Diener (1992); Watson and Tellegen (1985). p 0.001. Table 5. Empirical location of personality dimensions in the two-dimensional affective space via CIRCUM-extension Japanese sample (N = 450) English sample (N = 535) Angle Zeta VAF Angle Zeta VAF Neuroticism 176 0.31 97 176 0.47 98 Extraversion 24 0.21 84 28 0.22 84 Openness to Experience 63 0.07 32 0.10 86 Agreeableness 352 0.24 87 355 0.13 75 Conscientiousness 15 0.18 90 18 0.17 84 Angle refers to the estimated angular position of the personality dimension within the two-dimensional affective space. Zeta refers to the estimated communality index for the personality dimension and indicates the correlation between the personality dimension and the common score. Model fit for placing a personality dimension within the circumplex was assessed by the Variance Accounted for (VAF). Russell, Yik, and Steiger (in press). Value cannot be computed in CIRCUM-extension.

Momentary affect in the Japanese 91 axis is at least as close. From our perspective, it is the entire structure rather than on specific dimensions within it which is fundamental. Conclusion Figure 3. The correlation of affect variables with a personality dimension as a function of the angle within the circumplex for 14 affect variables. The value for the affect variable at 0 is repeated at 360 to show the complete sine wave. N, Neuroticism; E, Extraversion; O, Openness to Experience; A, Agreeableness; C, Conscientiousness. The angles obtained were not identical between the two samples. Meyer and Shack (1989) suggested that researchers could use personality dimensions to locate the fundamental dimensions of affect. Their prediction was that personality, or at least E and N, would fall at 45 and 135 in the space of Figure 2. Their prediction was borne out neither in Japanese nor in English. Deviations from these predictions were large, and the horizontal The present findings lend support to the viability of the structural model in Figures 1 and 2 as an integration of various dimensional models for momentary affect in both English and Japanese. To compare the Japanese results (Figure 2) with those in English (Figure 1), one can simply superimpose the two figures on top of each other. It becomes immediately obvious how similar the empiric placements of the affect variables are. Indeed, the rank order of the 14 variables is almost identical in English and Japanese (rank order correlation = 0.98). Further, the locations of the variables agree very well with the original authors conceptualizations: For instance, Thayer (1996) defined his Energy scale as pleasant activation and it indeed fell in the pleasant activated quadrant in Figure 2. The present study adopted the imposed-etic approach (Berry, 1969) in which translations of scales originating from English were administered to a sample of Japanese respondents. This approach emphasizes similarities across cultures and can be blind to indigenous constructs or processes. Given the richness of the emotion lexicon of Japanese, the possibility remains that additional affect dimensions would emerge with more indigenous items. Results obtained in the present study represent a first step towards studying affect and its external correlates in the Japanese culture. Here affect was studied at a broad general level high in the affect hierarchy, and further studies are much needed to examine more specific affective dimensions at a lower level in a hierarchy. We suspect that cultural differences will be obvious the lower one goes in that hierarchy. In addition to studying the structure of affect and its relation to personality, there are a huge number of questions about affect. One of our goals in the present study was therefore to provide ready-to-use, psychometrically sound affect measures for use in Japanese. We end

92 M. S. M. Yik, J. A. Russell and N. Suzuki here by noting that these scales provide a brief and efficient means of capturing affect found for Japanese-speaking respondents. Completing all 44 scales takes about 25 min. In basic research on structural relations between affect and other psychological variables in which measurement errors must be minimized, we recommend all 44 scales. In many research problems, it would be more practical to use one of the three response formats. If even more brevity is required, the four scales for Pleasant, Unpleasant, Activated, and Deactivated affect suffice to predict scores on the remaining dimensions. So, these scales can be used for a great variety of purposes with minimal cost. References Berry, J. W. (1969). On cross-cultural comparability. International Journal of Psychology, 4, 119 128. Browne, M. W. (1992). Circumplex models for correlation matrices. Psychometrika, 57, 469 497. Costa, P. T. Jr, & McCrae, R. R. (1984). Personality as a lifelong determinant of well-being. In C. Malastesta & C. Izard (Eds.), Affective processes in adult development and aging (pp. 141 157). Beverly Hills, CA: Sage. Costa, P. T., Jr, & McCrae, R. R. (1992). The Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-Factor Inventory (NEO-FFI) professional manual. Odessa, FL: Psychological Assessment Resources. Costa, P. T., Jr, & McCrae, R. R. (1995). Solid ground in the wetlands of personality: A reply to Block. Psychological Bulletin, 117, 216 220. Diener, E. (1984). Subjective well-being. Psychological Bulletin, 95, 542 575. Feldman Barrett, L., & Russell, J. A. (1998). Independence and bipolarity in the structure of current affect. Journal of Personality and Social Psychology, 74, 967 984. Larsen, R. J., & Diener, E. (1992). Promises and problems with the circumplex model of emotion. In M. S. Clark (Ed.), Review of Personality and Social Psychology: Emotion (Vol. 13, pp. 25 59). Newbury Park, CA: Sage. Larsen, R., & Ketelaar, T. (1991). Personality and susceptibility to positive and negative affect. Journal of Personality and Social Psychology, 61, 132 140. McCrae, R. R., & Costa, P. T. Jr (1991). Adding Liebe und Arbeit: The full five-factor model and well-being. Personality and Social Psychology Bulletin, 17, 227 232. Mehrabian, A., & Russell, J. A. (1974). An approach to environmental psychology. Cambridge, MA: MIT Press. Meyer, G. J., & Shack, J. R. (1989). Structural convergence of mood and personality: Evidence for old and new directions. Journal of Personality and Social Psychology, 57, 691 706. Ogawa, T., Monchi, R., Kikuya, M., & Suzuki, N. (2000). Development of the general affect scales. Japanese Journal of Psychology, 71, 241 246. Ogawa, T., & Suzuki, N. (2000). Emotion space as a predictor of binocular rivalry. Perceptual and Motor Skills, 90, 291 298. Ogawa, T., Takehara, T., Monchi, R., Fukui, Y., & Suzuki, N. (1999). Emotion space under conditions of perceptual ambiguity. Perceptual and Motor Skills, 88, 1379 1383. Remington, N. A., Fabrigar, L. R., & Visser, P. S. (2000). Reexamining the circumplex model of affect. Journal of Personality and Social Psychology, 79, 286 300. Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39, 1161 1178. Russell, J. A., Yik, M. S. M., & Steiger, J. H. (in press). A 12-point circumplex model of affect. Manuscript, Submitted for Publication. Shimonaka, Y., Nakazato, K., Gondo, Y., & Takayama, M. (1999). Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-Factor Inventory (NEO-FFI) manual for the Japanese version Tokyo: Tokyo Shinri. (In Japanese.) Tellegen, A. (1985). Structures of mood and personality and their relevance to assessing anxiety, with an emphasis on self-report. In A. H. Tuma, & J. D. Maser (Eds.), Anxiety and anxiety disorders (pp. 681 706). Hillsdale, NJ: Erlbaum. Thayer, R. E. (1996). The origin of everyday moods: Managing energy, tension, and stress. New York: Oxford University Press. Watson, D., & Clark, L. A. (1984). Negative affectivity: The disposition to experience aversive emotional states. Psychological Bulletin, 96, 465 490. Watson, D., & Clark, L. A. (1992). On traits and temperament: General and specific factors of emotional experience and their relation to the Five-Factor Model. Journal of Personality, 60, 441 476. Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54, 1063 1070.

Momentary affect in the Japanese 93 Watson, D., & Tellegen, A. (1985). Toward a consensual structure of mood. Psychological Bulletin, 98, 219 235. Yik, M. S. M. (1998). A circumplex model of affect and its relation to personality: A five-language study. Unpublished doctoral Dissertation, Vancouver, British Columbia: University of British Columbia. Yik, M. S. M., Russell, J. A., & Feldman Barrett, L. (1999). Structure of current affect: Integration and beyond. Journal of Personality and Social Psychology, 77, 600 619. (Received November 29, 2000; accepted May 17, 2002)