Perceived Stress Scale

History

Before the PSS, stress assessment focused on objective indicators of specific stressors, such as frequencies of chronic illness or family loss, overlooking subjective interpretation. Cohen et al. (1983) viewed this void as an unwanted quality and developed the PSS based on Lazarus's original transactional model of stress, which argues that the experience of a stressor is influenced by evaluations of how well one can manage it given coping resources. The original PSS consists of 14 items purported to form a unidimensional scale of global perceived stress.

Although scores on the 14-item PSS tend to exhibit good reliability estimates across the literature, four items perform poorly when evaluated using exploratory factor analysis. As a result, the PSS is commonly implemented using the 10-item form. Cohen et al. (1988) further reduced the PSS to a four-item form for quick measurements; however, scores on the 4-item PSS tend to exhibit lower reliability estimates than researchers would like.

Factor structure

Although Cohen et al. (1983) originally argued the PSS to be a unidimensional measure of perceived stress, the research community generally views the 14- and 10-item forms as two dimensional, consisting of positively and negatively phrased items. Under exploratory factor analysis, the negatively phrased items load onto a second factor separate from the positively phrased items, giving the appearance of a method effect. However, inspection of the items suggests a substantive difference: negatively phrased items on their face characterize perceived helplessness, while positively phrased items may characterize perceived self-efficacy. Some support for this conceptualization has emerged.

For example, Hewitt et al. (1992) found that the perceived helplessness items tended to predict depression in both men and women, while the positively phrased items tended to predict depression in women only. This pattern of differential prediction between genders indicates the negatively phrased items are tapping something substantive rather than a method effect. Although the preponderance of evidence is on the side of a multidimensional measure, the issue has not been fully resolved owing in part to limitations in the methodological knowledge available on distinguishing between substantive and method variance.

Reliability

Across diverse conditions, researchers report relatively satisfactory reliability estimates for scores on the 14- and 10-item forms. For example, Roberti et al. (2006) reported reliability estimates of .85 and .82 in a university sample for scores on the perceived helplessness and perceived self-efficacy scales, respectively. Highly similar results were reported in Taylor (2015) in a sample of middle-aged adults.

However, one limitation for much of the reliability estimates reported in the extant literature is the overly restrictive requirement of tau-equivalence for accurate reliability estimates. Tau-equivalence requires each item of a scale to have approximately the same size of relationship with the unobserved characteristic driving responses, as well as consistent reliability regardless of the latent level. Therefore, it is largely unknown whether items on the PSS assign scores with the same degree of reliability for respondents with high latent levels as respondents with low latent levels.

One exception is in Taylor (2015), where the graded response model was utilized to study reliability levels across levels of the two latent variables. Taylor (2015) reported that the perceived self-efficacy and perceived helplessness subscales tended to estimate scores reliably across levels of the latent variables except among respondents with exceptionally low levels of perceived helplessness and exceptionally high levels of perceived self-efficacy. Finally, much less is known about the reliability of scores from the 4-item form.