The relationship between positive and negative affect ...

JOURNAL OF

Journal of Research in Personality 36 (2002) 463–475

RESEARCH IN PERSONALITY www.academicpress.com

The relationship between positive and negative affect in the Positive and Negative Affect Schedule Stefan C. Schmukle,a,* Boris Egloff,a and Lawrence R. Burnsb a

Department of Psychology, Johannes Gutenberg-Universit€at Mainz, D-55099 Mainz, Germany b Department of Psychology, Grand Valley State University, Allendale, MI, USA

Abstract The Positive and Negative Affect Schedule (PANAS; Watson, Clark, & Tellegen, 1988) is one of the most widely used affect scales. Nevertheless, the relation between its two scales, positive affect (PA) and negative affect (NA), is still controversial. Previous results that suggest independence between NA and PA were limited to manifest variables. In this study, the relation between PA and NA for both state and trait instructions was analyzed by means of structural equation modeling. Two hundred ninety-two participants responded to the PANAS at three occasions of measurement. No association was found between trait PA and NA, but significant negative correlations between state PA and NA emerged. In the second step, the observed variance of state PA and NA was decomposed into a dispositional component, an occasionspecific component, a method-specific component, and a component due to measurement error by employing a multi-construct latent state–trait model. This analysis confirmed and extended the results of our first analysis: the dispositional components of state PA and NA were unrelated. In contrast, the situation-specific components were negatively associated. Thus, the negative correlation between state PA and NA could be traced back to situation-specific effects. Ó 2002 Elsevier Science (USA). All rights reserved. Keywords: Positive affect; Negative affect; PANAS; Structural equation modeling; Latent state– trait analysis

*

Corresponding author. Fax: +49-6131-39-22483. E-mail address: [email protected] (S.C. Schmukle).

0092-6566/02/$ - see front matter Ó 2002 Elsevier Science (USA). All rights reserved. PII: S 0 0 9 2 - 6 5 6 6 ( 0 2 ) 0 0 0 0 7 - 7

464

S.C. Schmukle et al. / Journal of Research in Personality 36 (2002) 463–475

1. Introduction The Positive and Negative Affect Schedule (PANAS; Watson, Clark, & Tellegen, 1988) is certainly one of the most widely used affect scales. It has been successfully employed in a variety of studies, concerning, for instance, perceived stress and health complaints (Watson & Pennebaker, 1989), social activities (Watson, Clark, McIntyre, & Hamaker, 1992), anxiety and depressive disorders (Watson, Clark, & Carey, 1988), job-search behavior (Burger & Caldwell, 2000), and cigarette consumption (Becona, Vasquez, Fuentes, & Lorenzo, 1999). The PANAS scales have been translated into different languages (see, e.g., Krohne, Egloff, Kohlmann, & Tausch, 1996, for a German version, and Sandin et al., 1999, for a Spanish version) and have been adapted for children (PANAS-C; Laurent et al., 1999). Although the PANAS has been extensively used for over 10 years, its structure, that is, the relation between its two scales, positive affect (PA) and negative affect (NA), is still controversial. Watson et al. (1988) as well as Watson and Clark (1997) showed that the correlation between positive and negative affect is very low and stable across different time frames when using the PANAS. It should be stressed, however, that these analyses were limited to manifest variables. Green, Goldman, and Salovey (1993) showed that measurement error may lead to an underestimation of the correlation between positive and negative moods. Consequently, Tellegen, Watson, and Clark (1999) could show that the correlation between the two PANAS scales was actually higher when taking measurement error into account. However, this correlation was still low enough to suggest relative independence. In this study, time frame was not varied as participants consistently indicated how they felt today. The magnitude of the relationship between PA and NA could also depend on the time frame: Diener and Emmons (1984) showed that measures of momentary affect suggest higher correlations between PA and NA than measures of affect that capture longer time periods. Although this study used other affect descriptors than the PANAS items—and the relation between PA and NA certainly depends on the affect measure (Egloff, 1998)—higher correlations might especially be found for the state PANAS when measurement error is considered. Unfortunately, an investigation of the structure of the PANAS for different time intervals by using latent variables has not yet been published. In our first analysis, we aim at filling this gap by analyzing the relation between the latent variables PA and NA given two different instructions, a state and a trait one. In the second part, the state PANAS is further analyzed. Watson et al. (1988) showed that even state PA and NA exhibited a significant level of stability (r :50, time interval: 8 weeks). This result reflects the strong dispositional component of affect. That is, even momentary moods are, to a certain extent, reflections of one’s general affective level. By using latent


465

state–trait analysis (Steyer, Schmitt, & Eid, 1999), it is possible to separate occasion-specific effects, dispositional effects, and measurement error of these ratings and to compute the size of these components. We expected the dispositional component of the state PANAS to show the same structure as the trait PANAS. In addition, we analyzed the structure of the occasionspecific component of the state PANAS, that is, the structure of the state PANAS without its dispositional component. According to the results of Diener and Emmons (1984), the negative correlation between NA and PA should be higher for this occasion-specific component.

2. Method Participants were 292 students (205 women and 87 men) of Grand Valley State University, Allendale, USA with a mean age of M ¼ 21:49 years (SD ¼ 4:64 years). There were three occasions of measurement with a time lag of 1 week. On each occasion of measurement, participants were given the state PANAS, in which they were asked to indicate ‘‘how do you feel right now, that is, at the present moment.’’ Additionally, at the first occasion of measurement, participants responded to the trait PANAS (‘‘how do you feel in general, that is, on the average’’). The PANAS consists of 20 adjectives that are rated on a 5-point unipolar response scale, with 1 anchored with very slightly or not at all, 2 anchored with a little, 3 with moderately, 4 with quite a bit, and 5 with extremely. In our first analysis, the latent correlations between PA and NA for the state and trait PANAS were analyzed. To separate measurement error from true interindividual differences, the two scales had to be split into equivalent test halves each consisting of five items. The first test half of the PA scale consisted of alert, active, excited, determined, and proud and the second half consisted of attentive, inspired, enthusiastic, interested, and strong. The NA scale was split into afraid, upset, guilty, jittery, and distressed as the first test half and nervous, hostile, ashamed, irritable, and scared as the second one. The items were assigned to the test scales due to their content, mean, and standard deviation. These test halves were the manifest measures of the latent variables PA and NA, as can be seen in Fig. 1. To ensure empirical

Fig. 1. Model for analyzing the correlation between the latent variables positive affect (PA) and negative affect (NA) with two test halves at one occasion of measurement.

466


identification, factor loadings of each factor were set to be equal. Then the correlation between these latent variables—separately for state and trait affect—was computed and tested for significance. In the second part of the analysis, the magnitude of the correlation between NA and PA was analyzed for the dispositional and occasion-specific components of the state PANAS. These components can be separated by means of longitudinal confirmatory factor analysis (Marsh & Grayson, 1994) or latent state–trait analysis (Steyer, Ferring, & Schmitt, 1992; Steyer et al., 1999). These models have been successfully applied in different areas of psychology (e.g., Eid & Diener, 1999; Eid, Schneider, & Schwenkmezger, 1999; Schmitt, 2000; Steyer, Majcen, Schwenkmezger, & Buchner, 1989).1 Fig. 2 shows the latent state–trait model for two test halves and three occasions of measurement.2 The person-specific latent trait variables P1 and P2 , which are measured by the manifest variables of one specific test half at all the three occasions (e.g., Y11 , Y12 , and Y13 for the first test half), assess stable interindividual differences. These variables characterize a person across different occasions of measurement for each test half. The latent occasion-specific variables O1 , O2 , and O3 , which are measured by both test halves at one specific occasion (e.g., Y11 and Y21 for the first occasion), assess the occasionspecific deviations of the momentary states from the stable person-specific components, thus, occasion-specific interindividual differences. Expressed more formally, this means that the observed value of the ith test half at the kth occasion of measurement Yik is a linear combination of a person-specific latent trait variable Pi for each test half i, a latent variable measuring the influences of occasion-specific effects Ok , an error variable Eik , and a constant aik . It is further assumed that the loadings of the manifest variables on the latent variables are all equal to one 1

For an extensive introduction to latent state–trait analysis the interested reader is referred to Steyer et al. (1999). 2 This specific latent state–trait model is one of the four different longitudinal factor analysis models that was proposed by Marsh and Grayson (1994). The four models differ in two points: (1) either the test-specific or the occasion-specific latent variables were assumed to covary and (2) either the latent variables were correlated (first-order model) or a higher-order factor was used to explain the correlation between the first-order factors (higher-order model). Concerning (1), we decided to use a model with the assumption of covarying test-specific factors rather than covarying occasion-specific factors, because this is theoretically more plausible (Eid, 1996). Concerning (2) we decided to use a higher-order model because this provides the opportunity of conducting a multi-construct analysis. A further alternative model was proposed by Eid et al. (1999). It has a general factor on which all observed variables load and a method factor on which only the observed variables of one test half load. This model has theoretical advantages because a second-order factor is not necessary (Eid, 1996). We decided not to use this model because (1) the method specificity can only be computed for one test half and (2) this model leads to different results for the relation between the different constructs depending on the test half the method factor refers to. Nevertheless, we also analyzed this model and found similar results.


467

Fig. 2. A latent state–trait model for two test halves and three occasions of measurement. Yik ¼ observed score for test half i at occasion of measurement k; P ¼ general latent trait variable; Pi ¼ test-specific latent trait variables; Ok ¼ occasion-specific factors; Mi ¼ method factors; Eik ¼ error variables; ci ¼ loading parameters.

Yik ¼ aik þ Pi þ Ok þ Eik :

ð1Þ

The two latent trait variables P1 and P2 should correlate 1.00, if the two test halves are perfectly homogenous. But often they are not, and then different test halves do not only measure a common latent factor, called the general (i.e., test-half unspecific) latent trait variable P, but also a test-half-specific component, the so-called method factor Mi . Thus, the test-half-specific latent trait variable Pi is a linear combination of the weighted general latent trait variable P and the method factor Mi P i ¼ ci P þ M i :

ð2Þ

To ensure identification of this model, the variances of the two method factors M1 and M2 are constrained to be equal. Concerning the correlations

468


between the latent variables, it is assumed that the latent occasion-specific variables Ok are uncorrelated, indicating that there are no carry-over effects between occasions of measurement. Furthermore, it is postulated that the latent person-specific variables are uncorrelated with the latent occasionspecific variables and the error variables, that the latent occasion-specific variables are uncorrelated with the error variables, and that all correlations among the error variables are zero. Based on these assumed independencies between the latent variables, the variance of an observed variable can be additively decomposed into the variances of the latent variables by using Eqs. (1) and (2) VarðYik Þ ¼ VarðPi Þ þ VarðOk Þ þ VarðEik Þ ¼ c2i VarðP Þ þ VarðMi Þ þ VarðOk Þ þ VarðEik Þ:

ð3Þ

Based on this decomposition, the following coefficients can be defined: the occasion specificity SpeðYik Þ ¼

VarðOk Þ VarðYik Þ

ð4Þ

is the component of variance of an observed variable that is due to occasionspecific interindividual differences. The common consistency ComConðYik Þ ¼

c2i VarðP Þ VarðYik Þ

ð5Þ

is the degree of observed variance that is due to stable interindividual differences which are common to both test halves, i.e., the dispositional component. The method specificity MetSpeðYik Þ ¼

VarðMi Þ VarðYik Þ

ð6Þ

indicates the degree of variance that is due to test-half-specific stable interindividual differences. The reliability is the component of variance of an observed variable that is not due to measurement error. Because of the additivity of the variances of the latent variables, occasion specificity, common consistency, and method specificity add up to the reliability coefficient: RelðYik Þ ¼ 1

VarðEik Þ VarðYik Þ

¼ SpeðYik Þ þ ComConðYik Þ þ MetSpeðYik Þ:

ð7Þ

This latent state–trait model can easily be extended to a multi-construct model, which consists of two or more of these models that are related to different constructs. By employing such a multi-construct latent state–trait model, the relation between dispositional and occasion-specific components


469

of different constructs can be determined. We will use such a model to analyze the relation between trait and situation-specific aspects of state PA and NA. The different models were estimated with EQS, version 5.7 (Bentler, 1995). Evaluation of model fit was based on multiple criteria. The most often used criterion, the v2 likelihood ratio statistic, tests the hypothesis that the population covariance matrix matches the model-implied covariance matrix. There are, however, two problems associated with this statistic. The first problem is that the v2 statistic is sensitive to deviations from normality, which results in an overestimation of v2 for non-normal data. Analyses of our items showed that the NA items of the PANAS deviated moderately from normal distribution: their skewness was in the range of 1.28–1.88 and their kurtosis varied between 1.38 and 3.81 for the two different test halves at three occasions of measurement for the state instruction. Both statistics showed somewhat lower values for the trait instruction. The PA items, however, did not deviate from normality. Hu, Bentler, and Kano (1992) showed that a corrected v2 statistic, the Satorra–Bentler scaled v2 , leads to appropriate results in the case of non-normal data. For this reason, we employed the Satorra–Bentler scaled v2 in addition to the conventional v2 , as recommended by Curran, West, and Finch (1996). The second problem with the v2 test consists of the fact that the probability of rejecting any model increases as N increases, i.e., almost every model will be rejected if the sample size is large enough (cf. Bentler & Bonett, 1980). As a consequence, alternative measures of fit, so-called fit indices, are frequently used. Hu and Bentler (1998) evaluated the sensitivity to model misspecification of different fit indices in a large Monte Carlo study with different sample sizes, estimation methods, distributions, and model misspecifications. Best results were obtained with the maximum likelihood method. The authors found that the maximum likelihood-based standardized root mean square residual (SRMR) proved to be sensitive to models with misspecified factor covariances and the maximum likelihood-based comparative fit index (CFI) was sensitive to models with misspecified factor loadings. Following authors’ recommendations, we used maximum likelihood estimation and employed both the SRMR and the CFI indices in our study. With respect to fit criteria, Hu and Bentler (1999) observed that CFI values of .96 or above in combination with SRMR values of .09 or below indicated a good fit between the hypothesized model and the observed data.

3. Results Table 1 displays the results of the first analysis. For the state PANAS scales a model with correlated factors showed a good fit for all three occasions of measurement. The correlations between the latent variables PA and

470


Table 1 Model fit and correlations between the latent variables positive affect (PA) and negative affect (NA)

State First occasion of measurementa Second occasion of measurementa Third occasion of measurementa Trait Model with correlated factors Without correlated factors

df

v2

p

SB-v2

p

CFI

SRMR

r (PA–NA)

3

4.10

.25

4.20

.24

.999

.022

).16

3

6.80

.08

6.34

.08

.994

.028

).17

3

6.75

.08

5.11

.16

.996

.026

).27

3

2.77

.43

2.11

.55

1.00

.012

).10

4

5.14

.27

4.01

.40

.998

.052

Note. N ¼ 292. Maximum likelihood estimation. SB-v2 ¼ Satorra–Bentler scaled v2 ; CFI ¼ comparative fit index; SRMR ¼ standardized root mean squared residual. a With correlated factors. * p < :05. ** p < :01.

NA were small (rs between ).16 and ).27). Despite their low value, all three correlations were significant and models without correlated factors had to be rejected for all occasions of measurement. For the trait PANAS a different result emerged: the correlation between the latent variables was not significant and the model without the assumption of a correlation between the latent variables had a good fit. Furthermore, the v2 difference test between the models with and without correlated factors was not significant, v2diff ð1Þ ¼ 2:37, p > :10. As the sample size of N ¼ 292 is large enough to detect even rather small population effects with adequate power,3 the results on trait level suggest an independence between PA and NA. Hence, the correlation between the latent variables PA and NA is dependent on the time frame: while we found small negative correlations between state PA and NA, trait PA and NA seemed to be independent. As momentary mood has a strong dispositional component, our second analysis examined whether the independence between trait PA and NA could also be found for the dispositional component of state affect. This analysis was based on the multi-construct latent state–trait model depicted in Fig. 3. As described in the method section, this model separates a 3 The power of the v2 difference likelihood ratio test was determined with an analysis proposed by Satorra and Saris (1985). For an assumed population effect of q ¼ :15 between the latent variables NA and PA a noncentrality parameter of k ¼ 5:95 was estimated, which means that the power of the v2 difference test (v2crit ¼ 2:71, df ¼ 1, N ¼ 272) is approximately .79.


471

Fig. 3. Parameter estimates of the multi-construct latent state–trait model. PA ¼ positive affect; NA ¼ negative affect; O ¼ occasion-specific factor; M ¼ method factor. For the manifest variables, the first index relates to the test half, the second to the occasion of measurement, e.g., PA21 denotes the second test half of PA at the first occasion of measurement.

dispositional component, an occasion-specific component, a test-half-specific component, and a component due to measurement error. The method factors and measurement errors were not allowed to correlate between the two constructs PA and NA.4 However, the occasion-specific components of the two constructs at one occasion of measurement k, Ok , and Ok , were allowed to correlate, indicating that occasion-specific interindividual differences in NA and PA are associated on an occasion of measurement. On the other side, the dispositional components, i.e., the latent trait variables PA and NA, were assumed to be uncorrelated to test whether PA and NA were independent of the trait level. The model shown in Fig. 3 had a good fit, v2 ð51Þ ¼ 55:81, p ¼ :30, Satorra–Bentler scaled v2 ð51Þ ¼ 48:46, p ¼ :57, CFI ¼ .998, SRMR ¼ .046. The v2 difference test with a model that assumed correlated latent trait variables was not significant, v2diff ð1Þ ¼ 0:54, p > :25, showing that PA and NA were not significantly correlated on the trait level (estimated r ¼ :06). The

4 The current model is not identified under the assumption that the measurement errors associated with each split half scale are correlated within time (response format bias) and across time (response format biases that endure). Thus, response bias cannot be separated from other latent variances in this model.

472


Table 2 Coefficients of reliability, occasion specificity, common consistency, and method specificity for the test halves of positive affect (PA) and negative affect (NA) Reliability

Occasion specificity

Common consistency

Method specificity

1

2

3

1

2

3

1

2

3

1

2

3

Positive affect First test half Second test half

.86 .94

.86 .91

.91 .90

.35 .42

.37 .42

.42 .45

.47 .47

.46 .44

.46 .41

.04 .04

.03 .03 .04 .04

Negative affect First test half Second test half

.84 .94

.79 .84

.88 .94

.41 .50

.29 .34

.46 .54

.41 .41

.47 .46

.39 .37

.03 .04

.03 .03 .04 .03

Note. N ¼ 292. 1 ¼ first occasion of measurement, 2 ¼ second occasion of measurement, 3 ¼ third occasion of measurement.

occasion-specific components of PA and NA, however, were significantly correlated within a range from r ¼ :32 to ).51. The v2 difference test with a model without the assumption of correlated occasion-specific components was highly significant, v2diff ð3Þ ¼ 94:37, p < :001. Table 2 shows the coefficients of reliability, occasion specificity, common consistency, and method specificity for the test halves of PA and NA: reliability of the test halves was high and method specificity was low. Furthermore, occasion specificity and common consistency coefficients revealed that state PA and NA were influenced by occasion-specific and latent trait effects to approximately the same degree.

4. Discussion This study demonstrated that PA and NA, as measured by the PANAS, are independent on the trait level. This finding confirms and extends the results reported by Watson et al. (1988) and Watson and Clark (1997) because our study took measurement error into account by means of structural equation modeling. Independence of PA and NA was shown both for trait affect (instruction: ‘‘how do you feel in general, that is, on the average’’) as well as for the dispositional component of the state PANAS (instruction: ‘‘how do you feel right now, that is, at the present moment’’). In contrast, there were small but significant correlations between state PA and NA, as could be expected from previous results by Diener and Emmons (1984). The latent state PANAS variables were further decomposed into dispositional and occasion-specific components. We could show that while the dispositional component of PA and NA was independent, the occasion-specific components of the PA and NA scales were significantly correlated. Thus, the rather high correlations for the occasion-specific components were


473

responsible for the lower, but still significant correlations between state PA and NA. These substantial correlations indicate that, although PA and NA are not correlated on the trait level, interindividual differences in PA and NA that are caused by occasion-specific aspects are negatively related. This suggests that the specific situation tends to affect both state PA and NA in opposite ways, i.e., when state PA increases, state NA tends to decrease and vice versa. Thus, whereas occasion-specific components are negatively correlated, the relation between common consistency components might be adequately described by affect independence. Some limitations to our study should also be noted. Naturally occurring affect in neutral situations, such as our data, might possibly lead to an overestimation of the trait component of state affect as well as to an underestimation of the magnitude of the state PA–NA association. This issue could be systematically investigated by studies that elicit stronger affects, like, e.g., experimentally induced positive or negative affect (e.g., Egloff, 1998; Larsen & Ketelaar, 1991). Another possibility would be the use of diary methods to assess affect-laden episodes that occur in the natural environment of the participants (e.g., Egloff, Tausch, Kohlmann, & Krohne, 1995; Gable, Reis, & Elliot, 2000). By employing such approaches, the magnitude of the state PA–NA correlation in different situations could be determined. Additionally, possible changes in the structure of the PANAS across different times of one emotional episode could be examined. Two further limitations arise from the fact that we used only items of the PANAS. First, since all these items share a common response format, response bias could not be assessed separately. However, by using different response formats Schimmack, B€ ockenholt, and Reisenzein (2002) could demonstrate that the effect of response styles on affect ratings is often overestimated. Thus, although it was not able to control response bias in this study, our results were not invalidated by this fact. Second, when measured with other scales than the PANAS, PA and NA usually show higher correlations (Watson, 1988). For this reason, one should be cautious in generalizing our findings to PA and NA in general. It remains an open question if the dispositional components of PA and NA as measured by other affect scales are uncorrelated, as we could show for the PANAS. Nevertheless, in view of our and Diener and Emmons’s (1984) results we would expect a higher correlation for occasion-specific than for dispositional components of state PA and NA also for other affect scales. To conclude, this study demonstrated that using a multi-construct latent state–trait analysis provides new insights into the relation of positive and negative affect. But it is certainly worthwhile and necessary to analyze other scales of momentary affect with this method to assess the generalizability of the results.

474


Acknowledgments The authors thank Judith Kappesser for helpful comments on an earlier version of this article.

References Becona, E., Vasquez, F. L., Fuentes, M. J., & Lorenzo, M. d-C. (1999). Anxiety, affect, depression and cigarette consumption. Personality and Individual Differences, 26, 113– 119. Bentler, P. M. (1995). EQS strucural equations program manual.. Encino, CA: Multivariate Software. Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin, 88, 588–606. Burger, J. M., & Caldwell, D. F. (2000). Personality, social activities, job-search behavior and interview success: distinguishing between PANAS trait positive affect and NEO extraversion. Motivation and Emotion, 24, 51–62. Curran, P. J., West, S. G., & Finch, J. F. (1996). The robustness of test statistics to nonnormality and specification error in confirmatory factor analysis. Psychological Methods, 1, 16–29. Diener, E., & Emmons, R. A. (1984). The independence of positive and negative affect. Journal of Personality and Social Psychology, 47, 1105–1117. Egloff, B. (1998). The independence of positive and negative affect depends on the affect measure. Personality and Individual Differences, 25, 1101–1109. Egloff, B., Tausch, A., Kohlmann, C.-W., & Krohne, H. W. (1995). Relationships between time of day, day of the week, and positive mood: exploring the role of the mood measure. Motivation and Emotion, 19, 99–110. Eid, M. (1996). Longitudinal confirmatory factor analysis for polytomous item responses: model definition and model selection on the basis of stochastic measurement theory. Methods of Psychological Research Online, 1, 65–85. Available: www.mpr-online.de. Eid, M., & Diener, E. (1999). Intraindividual variability in affect: reliability, validity, and personality correlates. Journal of Personality and Social Psychology, 76, 662–676. Eid, M., Schneider, C., & Schwenkmezger, P. (1999). Do you feel better or worse? The validity of perceived deviations of mood states from mood traits. European Journal of Personality, 13, 283–306. Gable, S. L., Reis, H. T., & Elliot, A. J. (2000). Behavioral activation and inhibition in everyday life. Journal of Personality and Social Psychology, 78, 1135–1149. Green, D. P., Goldman, S. L., & Salovey, P. (1993). Measurement error masks bipolarity in affect ratings. Journal of Personality and Social Psychology, 64, 1029–1041. Hu, L., & Bentler, P. M. (1998). Fit indices in covariance structure modeling: sensitivity to underparameterized model misspecification. Psychological Methods, 3, 424–453. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55. Hu, L., Bentler, P. M., & Kano, Y. (1992). Can test statistics in covariance structure analysis be trusted? Psychological Bulletin, 112, 352–362. Krohne, H. W., Egloff, B., Kohlmann, C.-W., & Tausch, A. (1996). Untersuchungen mit einer deutschen Version der ‘‘Positive and Negative Affect Schedule’’ (PANAS) [Investigations with a German version of the PANAS]. Diagnostica, 42, 139–156. Larsen, R. J., & Ketelaar, T. (1991). Personality and susceptibility to positive and negative emotional states. Journal of Personality and Social Psychology, 61, 132–140.


475

Laurent, J., Catanzaro, S. J., Joiner, T. E., Rudolph, K. D., Potter, K. I., Lambert, S., Osborne, L., & Gathright, T. (1999). A measure of positive and negative affect for children: scale development and preliminary validation. Psychological Assessment, 11, 326–338. Marsh, H. W., & Grayson, D. (1994). Longitudinal confirmatory factor analysis: common, time-specific, item-specific, and residual-error components of variance. Structural Equation Modeling, 1, 116–145. Sandin, B., Chorot, P., Lostao, L., Joiner, T. E., Santed, M. A., & Valiente, R. M. (1999). Escalas PANAS de afecto positivo y negativo: validacion factorial y convergencia transcultural [The PANAS scales of positive and negative affect: factor analytic validation and cross-cultural convergence]. Psicothema, 11, 37–51. Satorra, A., & Saris, W. E. (1985). Power of the likelihood ratio test in covariance structure analysis. Psychometrika, 50, 83–90. Schimmack, U., B€ ockenholt, U., & Reisenzein, R. (2002). Response styles in affect ratings: making a mountain out of a molehill. Journal of Personality Assessment, 78, 461–483. Schmitt, M. (2000). Mother–daughter attachment and family cohesion. European Journal of Psychological Assessment, 16, 115–124. Steyer, R., Ferring, D., & Schmitt, M. J. (1992). States and traits in psychological assessment. European Journal of Psychological Assessment, 8, 79–98. Steyer, R., Majcen, A.-M., Schwenkmezger, P., & Buchner, A. (1989). A latent state–trait anxiety model and its application to determine consistency and specificity coefficients. Anxiety Research, 1, 281–299. Steyer, R., Schmitt, M., & Eid, M. (1999). Latent state–trait theory and research in personality and individual differences. European Journal of Personality, 13, 389–408. Tellegen, A., Watson, D., & Clark, L. A. (1999). On the dimensional and hierarchical structure of affect. Psychological Science, 10, 297–303. Watson, D. (1988). The vicissitudes of mood measurement: effects of varying descriptors, time frames, and response formats on measures of positive and negative affect. Journal of Personality and Social Psychology, 55, 128–141. Watson, D., & Clark, L. A. (1997). Measurement and mismeasurement of mood: recurrent and emergent issues. Journal of Personality Assessment, 68, 267–296. Watson, D., Clark, L. A., & Carey, G. (1988). Positive and negative affectivity and their relation to anxiety and depressive disorders. Journal of Abnormal Psychology, 97, 346–353. Watson, D., Clark, L. A., McIntyre, C. W., & Hamaker, S. (1992). Affect, personality, and social activity. Journal of Personality and Social Psychology, 63, 1011–1025. 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. Watson, D., & Pennebaker, J. W. (1989). Health complaints, stress, and distress: exploring the central role of negative affectivity. Psychological Review, 96, 234–254.