Positive affect (PA) and negative affect (NA) are the higher order dimensions that provide measures of emotional experience. Although their names might imply that they represent opposite poles of the same dimension, PA and NA are indicative of orthogonal dimensions of affective experience, with high-NA representing subjective distress and unpleasant engagement, and low-NA denoting calmness and serenity. On the other hand, PA is an expression of the extent to which individuals experience pleasurable engagement with the environment. Thus, emotions such as alertness and enthusiasm characterize high PA, while lethargy and sadness indicate low PA.¹

Watson and Clark¹ concluded that "in our own studies and in virtually all published self-report studies that we have subsequently reanalyzed, we have encountered the same two large, bipolar dimensions" p. 220. Also, Cacioppo and Gardner² concluded that independent biological mechanisms appear to regulate experiences of PA and NA (probably because PA and NA had different roles in evolution), one indication of which would be differential predictors of PA and NA elicitation.

PA and NA have associations with the tripartite model.³ According to this model, anxiety and depression share NA as a common factor, whereas depression is specifically characterized by low levels of PA, and anxiety by physiological hyperarousal (PH). This model explains the frequent comorbidity of anxiety disorders and depressive disorders by suggesting that both disorders share the common dimension of NA.⁴ The tripartite model has gained broad and growing acceptance, in adult population,⁵ in a child clinical sample⁶ and in elderly sample.⁷

The Positive and Negative Affect Schedule (PANAS) is one of the most widely utilized measures of PA and NA, and was developed by Watson et al.⁸ PANAS is two by 10-item tools designed to assess PA and NA. Each of the items is rated from 1 (very slightly or not at all) to 5 (very much).

Although many studies have shown that PANAS has good psychometric properties, some problems remain. Watson et al.⁸ insisted that PA and NA are independent, but findings about associations between PA and NA are inconsistent. A previous study in Caucasians showed either zero⁹ or negative correlations^8,10 between PA and NA, whereas a positive correlation between PA and NA was reported for the Japanese version of PANAS.¹¹ Thus, it is possible that associations between PA and NA vary across countries according to the notion of cultural affect diversity.¹²

In addition, confirmatory factor analysis (CFA) findings of the two-factor model of PANAS have been contradictory, whereas exploratory factor analysis (EFA) results yielded a robust two-factor solution. Some studies have produced adequate fits to the data with the CFA of PANAS, whereas others have observed poor fit. Thus, it remains to be determined whether the CFA technique provides a clear replication of Watson et al.⁸'s factor solution. Some researchers have examined the psychometric properties of PANAS in non-English speakers including Estonians,¹³ French,¹⁴ Germans,¹⁵ Japanese,¹¹ Russians,¹⁶ Spanish,¹⁷ Swedish,¹⁸ Turkish,¹⁹ Italian,²⁰ and Serbian.²¹

Recently, a Korean version of PANAS (K-PANAS)²² was developed, but it has some limitations. The first involved a translation problem which we found in the back-translation procedure. Lee et al.²² noted that the item, gi-min-han, which supposedly means "alert" has a meaning that is slightly different from the precise meaning of 'alert', i.e., 'are paying full attention to things around one'. We also found other problematic items, i.e., heung-mi-jin-jin-han (interested), heung-bun-doen (excited), dan-ho-han (determined), and ju-ui-gi-peun (attentive). Heung-mi-jin-jin-han is suitable for describing an object rather than an affect, whereas Heung-bun-doen can mean "agitated" or "excited". Dan-ho-han is suitable for describing a behavior rather than an affect, and has two different context-dependent meanings, i.e., "assertive" and "determined". Similarly, ju-ui-gi-peun has two meanings, i.e., "attentive" and "henhearted".

There is another problem in the study by Lee et al.²² concerning the factor analysis technique used. Four models of PANAS have been proposed: 1) an uni-dimensional model that loads all items onto one factor; 2) a multidimensional model composed of two interrelated factors (PA and NA); 3) a multidimensional model consisting of two independent factors (PA and NA); and 4) a hierarchical three-factor model composed of a single PA factor and NA as a second-order factor made up of two lower-order factor (PA, "upset", and "afraid").²³ Lee et al.²² conducted factor analysis using the EFA, but this technique is unsuitable for comparisons with other models.

The third problem with Lee et al.²²'s study concerns sample selection. In this study, undergraduate and community samples were combined, and factor structures could differ across these populations, which reduce model fit to the data. In addition, the sample included only non-clinical subjects, and therefore, we should be cautious about applying these findings to other groups, especially to clinical populations.

Because of the limitations and inconsistencies of previous studies, our aims in the present study were to; 1) determine the factor structure of K-PANAS among Koreans, 2) investigate the psychometric properties of K-PANAS in a clinical group, 3) verify the correlation between PA and NA among Koreans, and 4) to test whether the relationships of PA and NA with anxiety and depression support the tripartite model. We hypothesized that the correlation between PA and anxiety would be lower than that between PA and depression, and that PA would account for significant unique variance to the prediction of depression.

Our findings can be summarized as follows. 1) The reliability of the Korean version of PANAS is satisfactory. 2) The constructed models yielded an unacceptable fit for the data set. Results from our CFAs showed that neither model provided a good fit for the data. One possible explanation for the poor fit of the CFA models might be cultural differences. Thus, it is recognized that further refinements could be made by removing several items with insufficient loadings from each domain. However, it may be premature to do this, since further study with larger and more diverse samples might produce different results. 3) The modified model indicated good fit. 4) Moderate correlations between NA and the K-BAI and K-BDI provided strong evidence for convergent validity. However, contrary to prediction, the analysis showed a weak negative correlation between PA and the K-BDI and a weak positive correlation between PA and the K-BAI, which are at odds with the findings of a previous study,³⁸ which showed that NA correlated highly with the BDI (r=0.76) and the BAI (r=0.79), PA correlated moderately with the BAI (r=-0.38), and PA correlated highly with the BDI (r=-0.61). The weak association between depression and PA, found in the present study, contrasted with the significant and moderate association found by Jolly et al.³⁸ between the BDI and PA in their clinical sample. It appears that Koreans and Japanese do not to express their positive feelings freely, which restricts the PA score range, and thereby, may have attenuated associations with the K-BDI. The fact that the standard deviation of PA was smaller in our sample than in Jolly et al.³⁸'s sample (SD's=6.2 vs. 9.4) supports this explanation. 5) The regression analysis showed that variance unique to depression is predicted by PA. These findings are generally consistent with the tripartite model and suggest the usefulness of PA for differentiating depression and anxiety. 6) The association between PA and NA was positive, though weak, which is consistent with finding in the Japanese sample, but is contrary to results in Caucasians. Thus, finding could reflect differences between Asians and Caucasians in terms of the relationship between PA and NA.

Small but significant positive correlations between PA and NA are explained by the fact that the measure in this study assessed state PA and NA rather than trait PA and NA. Previous studies showed that while PA and NA were independent of each other on the trait level, the associations between PA and NA were significant on the state level.³⁹ These results suggest that whereas occasion-specific components of the PA and NA scales are positively correlated in the present analyses, the relation between the dispositional components of PA and NA could be different in the future study.

This study has four limitations. First, only self-reported mea-sures were included, and thus, relationships between variables may have been affected by questionnaire-specific method variance. Second, the sample size was inadequate. According to Guadagnoli and Velicer,⁴⁰ stability problems occur when sample sizes are small (n=50 to 300), and thus, the size of our patient sample (n=218) may have been insufficient to obtain a stable lower-order factor solution. Thus, factor analyses should be replicated with larger samples. Third, although the model modifications needed to yield a sufficient fit to the data were both conceptually and statistically upheld, cautious interpretation of these results is needed. Many methodologists have been criticized post hoc modifications which rely on chance, thus prevent the generalization of the findings.⁴¹ Finally, we did not assess the inter-rater agreement for the diagnoses which could have affected our results. Thus, additional analyses should be conducted to calculate the agreement coefficient.

K-PANAS seems to be a promising tool for assessing PA and NA, but would probably benefit from further refinement. Although our post-hoc CFA supported a two-factor solution, this does not imply that it does not require further confirmation. Because the hypothetical models did not adequately fit the sample, further study of this instrument is required to identify the best fitting model.⁴²