Validation of the 10-item Chinese perceived stress scale in elderly service workers: one-factor versus two-factor structure

Because of its extensive associations with health outcomes, stress has long been an important research topic. The transactional model regards stress as an interaction between the individual and the environment (Lazarus & Folkman 1984). Stress arises when one appraises an event as threatening to the accomplishment of important goals or overwhelming to one’s resources. The transactional meaning of stress thus not only incorporates environmental and personal characteristics, but also emphasizes the subjective appraisal of the event. The same stimulus may generate different interpretations, responses, and coping strategies among individuals with different experiences and personality traits.

In accordance with the transactional model of stress, Cohen, Kamarck, and Mermelstein (Cohen et al. 1983) developed the Perceived Stress Scale as a global stress measure. The scale assesses the degree to which situations are appraised as stressful during the previous month. Originally, this self-report scale comprised 14 items. Later the authors reported the shortened 10-item version (PSS-10) as psychometrically superior to the original 14-item version (Cohen & Williamson 1988). The PSS-10 has demonstrated adequate internal consistency, with Cronbach’s alpha coefficients ranging from 0.72 to 0.91 in previous studies conducted in samples of college students, participants of smoking-cessation program, adults in the community, workers of occupational health care centers, policewomen, medical students and hospital inpatients (Leung et al. 2010; Orucu & Demir 2009; Wongpakaran & Wongpakaran 2010). Previous research has also found evidence for the scale’s construct validity. It was negatively correlated with positive measures, such as perceived health in college students and participants of smoking-cessation program, and self-esteem of medical students and hospital inpatients. On the other hand it was positively correlated with negative measures, such as health complaints in college students (Otto et al. 2004), state anxiety and depressive symptoms in depression patients, medical students and hospital inpatients (Orucu & Demir 2009; Wongpakaran & Wongpakaran 2010), susceptibility to the common cold in healthy adults (Cohen et al. 1993), and emotional exhaustion in college students (Ramirez & Hernandez 2007).

Regarding the scale’s dimensionality, most researchers have found evidence for a two-factor structure (Eskin & Parr 1996; Wang et al. 2011; Wongpakaran & Wongpakaran 2010; Orucu & Demir 2009; Otto et al. 2004; Roberti et al. 2006; Reis et al. 2010). The two factors revealed in the EFAs were named Perceived Helplessness (comprised of items 1, 2, 3, 6, 9, and 10) and Perceived Self-efficacy (comprised of items 4, 5, 7, and 8, which are reversely coded when computing the total score).

In a study of 60 suicide survivors, Mitchell, Crane, and Kim (Mitchell et al. 2008) argued for a one-factor structure for PSS-10 even though the eigenvalue of the second factor was still bigger than 1. The main arguments were, firstly, the primary factor already accounted for sufficient amount of the total variance explained (56.6%), and secondly, being conceptualized as a global assessment of stress, a one-dimensional structure for PSS-10 is theoretically more coherent.

PSS, especially the 10-item version, has been widely adopted in health outcome studies. The PSS-10 has been translated and validated in Japanese (Mimura & Griffiths 2008), Swedish (Eskin & Parr 1996), Spanish (Ramirez & Hernandez 2007), Turkish (Otto et al. 2004), Portuguese (Reis et al. 2010), French (Lessage et al. 2012), and Thai (Wongpakaran & Wongpakaran 2010). Given the close relation between stress and a wide range of well-being measures such as perceived health (Cohen et al. 1983), negative affect (Cohen et al. 1993; Ho et al. 2004), and workplace well-being (Duran et al. 2006; Prosser et al. 1997; Ro et al. 2010), it is essential to have a measure of perceived stress validated in the Chinese people. Although the PSS-10 has been translated into Chinese (Lee & Crockett 1994) and the Chinese PSS-10 (CPSS-10) has been applied in various previous studies (Ho et al. 2004; Gao et al. 2009; Chung & Tang 2006), the psychometric properties of CPSS-10 have rarely been examined. These previous studies have merely reported the scale’s internal consistency (Cronbach’s alphas ranging from 0.74 to 0.82), and provided minimal information on its factorial and construct validity. The only exceptions were more recent studies by Leung (Leung et al. 2010) and Wang (Wang et al. 2011). Leung examined the factor structure, internal consistency and construct validity of the 4, 10 and 14-item versions of CPSS. Leung concluded that the 10-item version CPSS, with a two-factor structure, showed the best overall psychometric properties. Wang evaluated the factor structure of CPSS-10 and recommended a 2-factor structure. However, since these two studies were on very specific groups, cardiac patients who smoked and policewomen respectively, the transferability of the findings to other population groups is unsure.

Because of the high relevance of the notion of stress in workplace, it is essential to validate CPSS-10 in working adults. The current study evaluated the psychometric properties of CPSS-10 in elderly service workers in Hong Kong.

For the first split sample, the KMO measure was found to be 0.74, while Barlett’s test of sphericity was significant, with χ ²(45) = 1021.61, p < 0.01, thus fulfilling the prerequisites for conducting EFA. With regards to the dimensionality of the CPSS-10, the scree plot showed that the curve leveled off after the first two components, with eigenvalues of the two factors greater than 1 (2.92 and 2.00). These findings suggested a two-factor solution for the scale (Figure 1).

Table 3 displays the goodness-of-fit indices of the CFA models in the second split sample. The one-factor model showed a poor data fit, with χ ²/df = 10.09, CFI = 0.62, TLI = 0.52, RMSEA = 0.14, 90% C.I. of RMSEA = (0.12 - 0.15), and SRMR = 0.11. None of the fit indices matched the cutoff criterion. On the other hand, the two-factor model revealed in the EFA showed adequate data fit. The fit indices, given by χ ²/df = 2.85, CFI = 0.93, TLI = 0.90, RMSEA = 0.06, 90% C.I. of RMSEA = (0.05 - 0.08), and SRMR = 0.04, were all adequate in accordance with the cutoff criteria. The standardized regression coefficients, as shown in Figure 2, ranged from 0.49 to 0.68 for Factor 1 and from 0.42 to 0.68 for Factor 2.

Model	χ2	df	χ2/df	CFI	TLI	RMSEA	CI90% (RMSEA)	SRMR
1-factor model	353.07	35	10.09	0.62	0.52	0.14	0.12 - 0.15	0.11
2-factor model	96.86	34	2.85	0.93	0.90	0.06	0.05 - 0.08	0.04

In general, Factor 2 score showed correlations with the validation scales in the opposite direction of the total and Factor 1 scores. The Factor 2 score was negatively correlated with exhaustion (r = −0.16), cynicism (r = −0.18), and reduced efficacy (r = −0.25), and positively correlated with vigor (r = 0.26), dedication (r = 0.24), and absorption (r = 0.17). The correlation between Factor 2 score and perceived workload was insignificant.

The PSS-10 intends to be a global and generic scale that measures the degree to which life events are appraised as stressful. Although the Chinese translation of the PSS-10 has been available for over ten years (Lee & Crockett 1994), few studies have closely examined its psychometric properties in samples of Chinese working adults. The existing studies were conducted in a very specific samples, which were cardiac patients who smoked and policewomen (Leung et al. 2010; Wang et al. 2011). The transferability of the findings of these studies is unsure. To our best understanding, the current study is the first study to validate the CPSS-10 in Chinese workers of elderly service.

The EFA with the first split sample revealed an unambiguous two-factor solution with a decent portion of explained variance and a factor loading pattern comparable to results of previous studies (Hewitt et al. 1992; Mimura & Griffiths 2004). Factor 1 (Perceived Helplessness) comprised six negative items, and Factor 2 (Perceived Self-efficacy) comprised four positive items, matching the findings of recent factorial analytic studies on the PSS-10 (Roberti et al. 2006; Reis et al. 2010). The whole scale of the CPSS-10 and Factor 1 Perceived Helplessness showed acceptable levels of internal consistency. With only 4 items, Factor 2 Perceived Self-efficacy showed a relatively lower Cronbach’s alpha coefficient, indicating marginally acceptable consistency.

With the second split sample, CFA revealed a poor fit for the one-factor model suggested by Mitchell, Crane, and Kim (Mitchell et al. 2008). In contrast, the two-factor model revealed in the EFA with the first split sample showed a much better fit, with goodness-of-fit indices meeting all the cutoff criteria.

The current study revealed a weak correlation between Factor 1 and 2 (r = −0.08, p < 0.05), which is in contrary to previous findings where the two factors were negatively correlated at moderate magnitude. In the study by Leung and her associates (Leung et al. 2010) on 1,860 Chinese cardiac patients who smoked, the inter-factor correlation was revealed to be −0.57. In Wang’s study with 240 Chinese policewomen, the inter-factor correlation was revealed to be −0.47 (Wang et al. 2011). Given the large sample size (992) from multiple sites and high response rate (93%) of the current study, the low inter-factor correlation revealed might suggest a relative independence of the two factors among Chinese elderly service workers. A low, insignificant inter-factor correlation poses some important questions. Are the two factors measuring the same construct? Should a composite score be created for the construct using scores of two uncorrelated factors? Since such low correlation has not been reported in previous studies, the finding should be further tested in different samples, especially with care workers in different social service settings.

With respect to construct validity, we found the CPSS-10 to be associated with various dimensions or manifestations of work well-being as signaled by its convergent property with perceived workload, exhaustion, cynicism and reduced efficacy, and its divergent property with vigor and dedication. The direction of correlations of Factor 1 was generally in line with the total score, whereas Factor 2 showed correlation direction opposite to that of the total score. These results, consistent with research findings on job stress (Duran et al. 2006; Prosser et al. 1997), provide empirical support for the scale’s construct validity in the work context.

Some limitations of the current study must be discussed. The current scale validation study solely relied on self-reported measures. Future studies might incorporate other assessment modalities, such as behavioral appraisals and physiological markers of stress, so as to further substantiate the construct validity. Also, we scrutinized the CPSS-10 in a convenience sample of working adults in the elderly service field. Although there was substantial diversity in their age, marital status, and education levels, over 83% of them were women. This imbalanced gender composition might lower the sample representativeness and limit the generalization of the findings to other occupations where the male-to-female ratio differs substantially from the current study sample. We therefore recommend that further studies be undertaken to evaluate the CPSS-10 in samples with different gender composition. Even so, the large sample size (N = 992) and high response rate (93%) of the current study did lend support to the accuracy of the findings. Another notable strength of this study is that we performed scale validation through simultaneous use of EFA and CFA on random split samples. Previous studies have rarely adopted such a cross-validation approach, with the exception of one recent report (Reis et al. 2010).

In summary, the Chinese version of the PSS-10 (CPSS-10) demonstrated a stable two-factor structure consistent with the findings of most previous studies. It further displayed acceptable internal consistency and adequate evidence for construct validity, and we recommend its use in stress-related research in the Chinese population.