The factor analysis revealed that for the original PSS, all the items of negative experience loaded highly on the first factor. Although some items reflecting positive experience were also strongly related to the first factor, these items also had substantial correlation with the second factor. This pattern of factor structure is nearly identical to that identified by previous empirical research. Cohen and Williamson [21
] showed a two-factor structure for the PSS in a US sample (n = 2387) and each factor reflected positively or negatively phrased items. Subsequent studies confirmed this factor structure in psychiatric patients in Canada (n = 96) [24
] and in psychology students in Mexico (n = 365) [22
]. In the current study as well, the first factors can be labelled as "negative perception" and the second factor as "positive perception." As mentioned above, however, Items 7, 10 and 13 highly loaded on the first factor in spite of positive statements. This might be due to a relatively small sample size (n = 222) for factor analysing the 14-item scale. In the study by Hewitt et al. (n = 96) [24
], factor loading was not reported about Items 10, 12 and 13 which might be differently grouped in terms of factorial nature.
On the other hand, as for the PSS-J, all items stating positive attitude were highly related to the first factor and all items of negative attitude to the second factor. Thus, labels for these factors can be "positive perception" and "negative perception" respectively.
The variance explained by the factor is somewhat different between the two cohorts in the current study, and also from that found in these previous studies. The first largest factor explained 42.7% and the second factor 11.0% in the UK sample, and 23.8% and 18.8% respectively in the Japanese sample in the current study. Previously reported variance was: 25.9% accounted for by the first factor and 15.7% by the second factor in the US sample [21
]; 31.4% and 15.2% respectively in the Canadian sample [24
]; and 32.6% and 15.4% in the Mexican sample [22
]. Such dispersion might be derived from cultural differences and different sample size.
Comparing the factor loading of the PSS and the PSS-J, although the magnitude of each factor was different, items stating positive attitude were gathered in the "positive perception" factor and items of negative attitude were in the "negative perception" factor for both scales. Also, the factor congruence coefficient indicated satisfactory factor agreement between the PSS and the PSS-J. It can be seen that the PSS and PSS-J were similar in factor structure. Cronbach's alpha for each factor was high. This suggested that all factors were internally consistent. The equivalence between the PSS and PSS-J was supported through a similar factor structure and factor loading on items. However, as pointed out by Cohen and Williamson [21
] and González and Landero [22
], the distinction between the two factors is considered irrelevant and total scores obtained by summing responses to all 14 items should be used for the purpose of measuring perceived stress.
Regarding both the PSS and PSS-J, all the items except for one (Item 12 in the PSS; Item 4 in the PSS-J) indicated acceptable item-total correlations. Generally, items showing an item-total correlation of 0.3 or lower are considered to be dropped from the scale [27
]. However, a scale with an acceptable Cronbach's alpha may still have one or more items with low item-total correlations [27
]. Thus, the findings of the current study were indicative of sufficient item-total reliability for the PSS and PSS-J.
As described in the translation section, a number of efforts were made to produce a Japanese version of the PSS as equivalent as possible to the original scale. Using intimate couples of a native English speaker and a native Japanese speaker probably contributed greatly to addressing problems that are likely to occur when an instrument is translated into other languages. These would include conceptual problems such as differences in conceptualisation and behaviours associated with the construct of a scale and inappropriateness of item content, and linguistic problems such as erroneous literal translation and poor wording [19
]. Multiple forward and backward translations and verification of the equivalence between the source and final version by a multi-disciplinary reviewer panel would also resolve these issues. However such couples are likely to differ from the population in general. The translation might, therefore, be biased although professional translators and those who generate the original items on such scales are equally unlikely to represent the general population.
While the sample size was large, another limitation of this assessment is that the subjects consisted of undergraduate nursing and pharmacy students who were recruited from a single institution in each country. Also, they were predominantly female. The findings may, therefore, be influenced by stress characteristics unique to them such as gender, stressors as a consequence of actually being undergraduates or particular factors relating to the subject of study. Further equivalence assessment using a sample that is more representative of the population in general would ideally be conducted to overcome these limitations. Certainly it is important that research using this scale in new population assesses the factor structure and internal consistency for its own sample.