Fliege, Herbert PhD; Rose, Matthias MD; Arck, Petra MD; Walter, Otto B. MD; Kocalevent, Rueya-Daniela MA; Weber, Cora MD; Klapp, Burghard F. MD, PhD
Stressis a key concept in health research (1). Definitions have basically focused on two major components of stress: a) stressors in terms of environmental conditions, and b) the person's reaction to stress. Stress reactions have been further differentiated theoretically, for example, into perceptional processing and emotional response. An empirical study based on structural equation modeling techniques found that the experience of stress was best represented by a two-factorial construct of stress (2). Environmental conditions were one factor; stress appraisal and emotional response in combination comprised the second.
With regard to the measurement of stress, it has been much debated whether or not we should limit ourselves to measuring stressors in terms of objective conditions, such as major life events or cumulative minor stressors (eg, daily hassles), or if we should rather concentrate on the person's stress reactions, in terms of their stress appraisal or emotional response (3). Stress research has shown an inconsistent picture of the effects of life events or daily hassles on health. Empirical studies have shown many instances in which an experience of accumulated or chronic stress led to physical health problems, whereas more severe but acute and temporally more “contained” life events could not predict illness to the same extent (4–6). Obviously, the personal impact of life events cannot be determined before the event has actually occurred (7). Other approaches have shifted the focus from specific objective stressors to more chronic and subjective stress experience (8).
Stress definitions have become more strongly focused on the subjective reactions to external events or demands (9). In the revised stress measure “Hassles and Uplifts Scale” (10), for example, we see both an environmental and an appraisal measure of stress, because it assesses not only whether a hassle occurs but also the perception of its severity or intensity. Nevertheless, many researchers have gone further and called for the development of instruments for the assessment of stress focused primarily on the subjective perception of the individual (3,11–13).
Against this background, Levenstein et al. (14) published the “Perceived Stress Questionnaire” (PSQ) 10 years ago. It had been their aim to overcome some of the difficulties concerning the definition and measurement of stress by putting the focus on the individual's subjective perception and emotional response. With this aim in mind, item wordings were designed to represent the subjective perspective of the individual (“You feel… ”). The presented stress experiences were intended to be abstract enough to be applicable to adults of any age, stage of life, sex, or occupation, but at the same time interpretable as specific to a variety of real-life situations. For example, “you feel under pressure from deadlines” could refer to anything from a payment, to an oncoming birthday party, or to a grant proposal. Factorial analyses resulted in 7 dimensions (harassment, irritability, lack of joy, fatigue, worries, tension, and overload). The authors made no a priori distinction between presumed stressor and stress response items. Although stress reactions certainly predominate the content of the scales, the overload subscale (“too many things to do,” “too many decisions to make,” etc) seemed at least to reflect the perception of stressful environmental conditions.
Psychometric characteristics proved to be favorable. PSQ scores correlated moderately with Cohen's Perceived Stress Scale (described later in this section), anxiety (State-Trait-Anxiety Inventory), and depression (CES-D Depression Scale). As far as external validity was concerned, PSQ values were higher in asymptomatic ulcerative colitis patients with an inflamed rectal mucosa than in those with a normal-appearing rectum. By choosing only patients in clinical remission, confounds resulting from the distressing effects of symptoms were eliminated. Furthermore, the authors were able to predict adverse health outcomes by means of PSQ values in a prospective study (15).
It seemed, therefore, that the instrument was properly qualified for research on stress and illness. However, there were certain flaws that suggested to us that reconsideration and further development of the questionnaire should take place.
First, the original validation study relied on relatively small samples. The overall sample comprised 230 subjects. Another point of concern is the number of scales. Although the pattern of item loadings could be satisfactorily interpreted, a total of 7 scales drawn from the original 36 items, tested on 230 subjects, seems fairly high from a statistical point of view. In 4 of the 7 scales, all item loadings scored below 0.50. This might indicate that a 7-factor solution does not rely on a sufficiently robust statistical basis. Finally, the clinical samples originally consisted only of patients with gastroenterological diseases. In a Spanish study, the PSQ was administered to psychiatric patients, nursing students, and healthy adults (16). Another study yielded moderate overall PSQ scores for a Swedish population sample (17). There was some evidence for external validity in a Thai sample of patients with peptic ulcer disease (18). In our opinion, the PSQ's dimensional structure should be investigated in different clinical groups and further reference values should be established.
Because there are some alternative stress questionnaires available that are also based on a concept of stress as a subjective experience, we will briefly point out what distinguishes them from the PSQ.
The Perceived Stress Scale (PSS) (11) is probably the most widely accepted of these measurements of stress. This 14-item questionnaire asks the respondent how often certain experiences of stress occurred in the last month. Stress—as opposed to challenge—is believed to result from experienced overload with further emphasis on experienced unpredictability and uncontrollability of events. This implies that the existence of stress in a person is partly inferred from information on the person's experience of lack of control. The content of the items is nonspecific. Two items directly address “stress” or “hassles,” three refer to situations of overload, whereas nine items refer to uncontrollable, unmanageable, or unpredictable situations. Thus, the PSS focuses on a more cognitive appraisal of stress and the respondent's perceived control and coping capability. A total score is provided. No subscales are reported.
The Index of Clinical Stress (ICS) (19) consists of 25 items. These items are designed to indicate affective states involved in the stress reaction. Cognitive appraisals, physical signs, or behavioral reactions are not considered. The ICS consists of one homogeneous scale. Subscales are not provided. The questionnaire lacks external validation.
The recently published Stress Response Inventory (SRI) (20) consists of 39 items that comprehensively focus on cognitive, emotional, behavioral, and somatic stress responses. In addition to a total score, its subscales differentiate between depression, frustration, anger, aggression, tension, and somatization. It does not cover the individual's perception of external stressors or demands.
The Trier Inventory for the Assessment of Chronic Stress (TICS) (21) is a validated German questionnaire focusing on chronic stress. The 39 items are factor-analytically assigned to 6 scales: work overload, work discontent, social stress, lack of social recognition, worries, and intrusive memories. The emphasis is on work-related and other socially stressful environmental conditions. To our knowledge, no English version has been published.
In comparison to the aforementioned instruments, we suggest that the PSQ is most useful
1. when, from a conceptual point of view, perceived stress should be asked as directly as possible, without inferring it from control or coping appraisals;
2. when, in addition to an overall score, different facets of perceived stress are of interest;
3. when information is wanted, not only concerning the person's stress response, but also concerning the perception of external stressors.
The first aim of our study was to investigate the dimensional structure of the questionnaire on a larger sample drawn from a different cultural context. Because questionnaires that might be included in routine use should keep respondent burden as small as possible, we aimed to reduce the length of the PSQ. In the course of item reduction, the explanatory power of different scales was to be balanced. Finally, we wanted to provide normative values for different clinical groups and healthy adults.
Concerning external validation, we expected that a higher perceived stress level in women who had experienced a spontaneous abortion would be associated with a higher concentration of certain immune parameters considered to be mediating factors in triggering spontaneous abortions (22). With regard to the instrument's sensitivity, we expected higher stress levels in women after miscarriage and inpatient groups, especially those who were treated with somatoform and depressive symptoms, and lower stress levels in pregnant women and healthy adults.
We administered the original 30-item questionnaire to one sample of participants (N = 650) in order to explore the factorial structure and to reduce the length of the questionnaire on one set of data. We then administered only the resulting item-reduced version of the questionnaire with 20 items to another sample (N = 1808) to test for structural stability on a completely separate set of data.
The study included two samples that involved a total of 2,458 participants.
1. The first sample (N = 650) is composed of the following:
* 246 patients hospitalized in the Psychosomatic Medicine ward, that is, patients with mental or behavioral disorders associated with at least one complex of somatic complaints or illness (included are somatoform, affective, eating disorders, other “neurotic” disorders, and personality disorders, all according to ICD-10 F3 to F6; excluded are organic, addictive, or psychotic disorders according to ICD-10 F0 to F2) (77.6% female, 22.4% male; age 38.9 ± 15.4 years, range 17–79),
* 81 female patients after miscarriages of unexplained origin (age 30.2 ± 7.7, range 17–41),
* 74 women after regular delivery (age 30.2 ± 5.0, range 19–43), and
* 249 medical students in the 4th year (51.1% female, 48.9% male; age 24.6 ± 2.9, range 20–41).
* Initial results from this sample have been published in German (23).
2. The second sample (n = 1808) is composed of the following:
* 559 Psychosomatic Medicine outpatients (diagnoses as above) (63.9% female, 36.1% male; age 37.8 ± 15.3, range 18–72),
* 184 outpatients with tinnitus (46.6% female, 53.4% male; age 42.1 ± 12.7, range 28–70),
* 144 outpatients with inflammatory bowel diseases (54.7% female, 45.3% male; age 39.6 ± 14.2, range 22–67),
* 587 women in routine care at week 8 of pregnancy (age 29.6 ± 5.3, range 17–44), and
* 334 healthy adults (61.6% female, 38.4% male; age 45.3 ± 15.6, range 18–88) who were visitors to a well-frequented institution for public education. (We defined only those participants as “healthy” who declared that they did not have any chronic or acute disease, were not in constant medical treatment, and were not in permanent need of medication).
3. Sensitivity to change was tested in the following:
* in 91 of the abovementioned sample of 246 Psychosomatic Medicine inpatients who were treated 5 weeks or more, so that we could measure at admission and after 5 weeks; treatment included a combination of single and group psychotherapy, relaxation training, sports, and in some cases antidepressants; and
* in 46 tinnitus outpatients who were assessed before and after 10 weekly sessions of progressive muscle relaxation training (27).
All patient groups were recruited in routine care. The students were recruited at the end of a course. The healthy adults were recruited before or at some time during the event that they visited. All participants were told about the aims of the study and gave their informed consent to participate.
Levenstein et al. (14) developed the PSQ to assess perceived stressful situations and stress reactions on a mainly cognitive and to some degree emotional level. With regard to stressors, the aim was to assess the subjective experience of their quality as stressful.
The scale construction was based on classical test theory and was carried out by factor analyses. The final instrument comprises 30 items that fell on factor analysis into 7 scales (harassment, overload, irritability, lack of joy, fatigue, worries, tension). Respondents rate how often an item applies to them on a 4-point scale (1: almost never, 2: sometimes, 3: often, and 4: usually). The general form of the instruction asks “in general, in the last two years”, the recent form asks “during the last month” (both in (14)). The PSQ Index and the scale values are mean values that are calculated from the raw item scores and linearly transformed to values between 0 and 1. The instrument was originally validated in English-speaking and Italian-speaking samples of gastroenterological inpatients, outpatients, hospital employees, and students (overall N = 230).
We translated the questionnaire into German. A clinical psychologist and English native speaker who had no prior knowledge of the instrument translated it back into English. Deviations from the original were examined, and the German translation was optimized accordingly.
On the samples presented here, we applied the “general form” of the questionnaire. According to the authors, it “integrates an individual's stress ‘in the long run,' [and] may be a superior predictor of health status” (Levenstein et al., 1993, p. 30). To avoid problems resulting from varying or insufficient memory recall, we omitted the time span of “the last year or two.” So the respondent was only asked to rate how often an item applied “in general.”
For purposes of validation, we administered the short measure of quality of life by the World Health Organization (WHOQOL-Bref (24)) and the abovementioned Trier Inventory of Chronic Stress (TICS (21)) to part of the sample.
The time needed to complete the questionnaire was recorded for the sample of 559 Psychosomatic Medicine outpatients.
An exploratory principal component factor analysis of the 30-item questionnaire was performed on the data from the first sample using SPSS. Because it could be expected that factors were correlated, an oblique rotation (promax, power coefficient = 4) was conducted. The factors were defined and interpreted based on the factor pattern matrix. We also tested whether the original 7-factor solution could be replicated on the German samples.
The first rationale for item selection was to balance the explanatory power between the scales by attaining scales of (approximately) equal length. The second rationale was to maximize reliability of the resulting scales by selecting those items that showed the highest corrected item-scale-correlation (Table 1).
We tested for structural stability on the data from the second sample, where subjects were administered only those 20 items that had resulted from the item selection. We tested a structure of 4 factors by means of linear structural equation modeling (SEM, Program AmosTM 4.0), allowing for one latent stress construct to underlie all 4 factors (Figure 1). In addition, we tested a 3-factorial and a 2-factorial structure, also allowing for correlations between the factors. We tested the 4-factorial structure for dimensional stability across groups by multisample analyses (MSA) using SEM. We performed several different comparisons between ill and healthy samples, combined and separate (Table 2). Because we expected mean values to differ across groups, we added a mean structure to the MSA model. To examine whether the factors can be defined the same way in all groups, cross-group equality constraints were imposed on the factor loadings (one loading was fixed to 1 in all groups). The mean of the factor was fixed to 0 in one group and estimated freely in the other groups (one indicator intercept per factor was fixed to 1 in all groups). Because this analysis did not aim to test hypotheses about means, no other equality constraints across groups were imposed.
For purposes of the MSA, all factor loadings of the observed variables (items) on latent traits (factors) and all loadings of the primary factors on the superordinate factor (“stress reaction”) and the correlation between “demands” and “stress reaction” were assumed to be constant across groups.
To corroborate construct validity, we performed comparisons with a measure of quality of life (WHOQOL-Bref (24)) and with a questionnaire of chronic stress (TICS (21)) that had been applied in two partial samples.
To determine criterion validity, we tested for associations between stress scores and immunological parameters in women suffering from a spontaneous abortion (22). We took decidual tissue biopsies and determined the occurrence of CD56+-NK-cells, CD8+- and CD3+-T-cells, tryptase+-mast cells (TMC+) and tumor necrosis factor-alpha+-cells (TNF-α+) by immunohistochemistry (IHC).
All biopsies were fixed in 5% formalin and embedded in paraffin. We examined two to four different sections of tissue for each patient. To make sure the trophoblast had been in contact with maternal immunocompetent cells and could have been a target of rejection, we stained the tissue with a monoclonal antibody against pancytokeratin (CK) to test for invasive fetal cells. Consecutive slides were stained with monoclonal antibody against mast cell tryptase, CD3, CD8, or CD56, respectively. Probes for human TNF-α mRNA were stored at −70°C until use. Five-micron paraffin sections were dewaxed and rehydrated, washed in DEPC-treated water, and immersed in 0.1N HCl followed by 2× SSC at RT. Sections were exposed to 10 μg/ml proteinase K and postfixed in 0.4% paraformaldehyde at 4°C. Hybridization was carried out at 59°C using S35 UTP-labeled cRNA. Afterward, sections were washed in 4× SSC and treated with RNase A (20 μl/ml). The slides were desalted, dehydrated, air dried, dipped into autoradiography emulsion, and developed. The sections were counterstained with hemalaun. Microscopic investigators were blinded to the patient's stress scores. The number of positive cells per square millimeter tissue was evaluated by two independent observers.
To examine sensitivity, we tested patient samples, pregnant women, and healthy adults for differences in their stress levels. All differences between samples were investigated by analysis of variance and secured by post-hoc t tests.
Exploration and Item Reduction
The Kaiser-Meyer-Olkin measure of the quality of the correlation matrix was high (KMO = 0.96). A significant Bartlett test of sphericity justified a dimension reducing procedure such as the factor analysis. The measure of sampling adequacy was over 0.80, so the items could be considered apt for factor analyses.
Exploratory analyses of all 30 items yielded a different solution from the original one (14). A forced 7-factor solution did not yield the original structure. Four factors were extracted with eigenvalues greater than 1. The eigenvalues' course was 12.5, 2.4, 1.5, and 1.0, then 0.9, 0.8, and 0.8, indicating a strong primary factor with 1 to 3 additional factors. We tested solutions with 4, 3 and 2 factors, respectively.
The 4-factor solution accounted for 58% of the variance. Item 03 (“You are irritable or grouchy”) did not load distinctly and item 11 (“You have too many decisions to make”) had a low communality (<0.50). They were therefore excluded. Item 17 (“you feel safe and protected”) loaded on factor III (−0.410) but also on factor I (−0.400). Still, we decided to accept this flaw and keep the item with a view to keeping scales of even length and in light of its satisfactory communality (0.58).
All remaining factor loadings were greater than 0.50 and the item's share of communality concerning one factor was at least 20% higher than its share of communality concerning any other factor. Communality varied between 0.50 and 0.71 around a mean of 0.60. See Table 1 for factorial solution, loadings, and item parameters.
The 3-factorial solution conformed to a simple factor structure except for items 03 and 17. It replicated factor I and factor IV of the 4-factorial solution, but factor II and factor III of the 4-factorial solution fell together on one factor. In the 2-factorial solution, the second factor replicated factor IV of the 4-factorial solution with only the addition of item 28. All other items loaded on a strong first factor.
We considered the 4-factorial solution the most informative one. The 3-factorial solution would have meant abandoning a consistently positively worded scale (factor III of the 4-factorial solution). As regards content, we considered a positively worded scale as advantageous, so we wanted to keep it, given sufficient structural stability. The 2-factorial solution seemed to replicate a theoretical distinction of the stress construct into perceived stressor (factor II) and stress reaction (factor I). We ultimately decided to investigate the 4-factorial solution more thoroughly and to include all three solutions in the confirmatory analyses.
We then selected those 5 items of each scale that showed the highest corrected item-scale-correlation (Table 1). Thus, a 20-item questionnaire of 4 scales with 5 items each resulted.
Scale 1 (worries) covers worries, anxious concern for the future, and feelings of desperation and frustration.
Scale 2 (tension) explores tense disquietude, exhaustion, and the lack of relaxation.
Scale 3 (joy) is concerned with positive feelings of challenge, joy, energy, and security. Because all items of this scale are positively worded, we opted for a positive name.
Scale 4 (demands) covers perceived environmental demands, such as lack of time, pressure, and overload.
An overall index score is calculated from all items, and linearly transformed to values between 0 and 1. For this purpose, the scale “joy,” which is positively coded, will be inversed. A high overall PSQ score means a high level of perceived stress.
Although all PSQ scales intercorrelate fairly highly, “demands,” which focuses on external stressors, shows the lowest correlations to the other three scales, which focus on the stress reaction (Table 3).
Confirmatory analyses, validation, and usability testing were all performed on the resulting 20-item questionnaire.
Following the above cited theoretical concepts (2), SEM was constructed as shown in Figure 1. Thus, the tested 4-factor solution specifies an additional latent variable “stress reaction” loading on the first 3 factors (worries, tension, joy) and covarying with “demands.” This resulted in a significant likelihood-ratio χ2 test (Table 2) with a global fit index (GFI) below 0.95 and an adjusted GFI below 0.90. However, because Hoelter's critical number (here 176) is considerably smaller than the sample size, any model would inevitably have been rejected applying those indices. Thus, we followed a recommendation to judge a model by a number of different criteria (25). The root mean squared residual below 0.05 is a criterion in favor of the model fit. Furthermore, the Tucker-Lewis index (TLI) and the comparative fit index (CFI) reached good values (≥0.90). Both are independent of sample size and either take into account model complexity (TLI) or model misspecification (CFI). Finally, a value of about 0.08 or less for the root mean standard error of approximation is considered to indicate a reasonable fit (26). This index allows for discrepancies between sample and population. Taking all this into account, we consider the model fit satisfactory.
Only the path weight between item 29 and “demands” fails to satisfy (0.47). This might be due to a positive item wording. A tentative exclusion of the item does not result in a closer model fit. The lowest path weight results for the overall sample, whereas in the subgroups this weight varies between 0.54 and 0.60. Because this item had a high loading in the original exploratory factor solution (0.72) and seems unproblematic as to content, we decided to keep it.
Multisample analyses yield that there is no appreciable gain in model fit by omitting the restriction of structural equality between groups. In sum, they confirm the assumption of a comparable dimensional structure in different samples.
Cronbach's alpha and split-half reliability values of the scales in the subgroups are all at least 0.70, in half the cases at least 0.80. Cronbach's alpha of the overall score is at least 0.85 and reliability at least 0.80 (Table 4).
Stress scales and overall score are negatively correlated, and the joy scale is positively correlated with quality of life (QoL) dimensions (p < .001) (Table 3). All PSQ scales correlate more highly with the psychological domain of the WHO-QOL than with other WHO-QOL domains. The correlational pattern with the TICS is altogether consistent with expectation. Five of the 6 TICS subscales are most highly correlated with the same PSQ scale (worries), whereas the TICS work overload scale is most strongly related to the PSQ demands scale.
Comparison to the Original
In the 650 subjects who completed the full questionnaire, the correlation between the 30-item overall score and the 20-item overall score was high (r = 0.95, p ≤ .001). To examine whether the level of the overall score and its measurement consistency were maintained in spite of the item reduction, we compared the gastroenterological sample of the original study (including many ulcerative colitis patients) with the inflammatory bowel disease (IBD) sample of the present study. Internal consistency of the original and the revised version is identical (α = 0.90). Mean values and distribution of the overall stress score of the original (0.42 ± 0.15, range 0.11–0.83) and the revised questionnaire (0.43 ± 0.17, range 0.02–0.87) do not differ. Mean values and distribution of the overall stress score of healthy adults in a Spanish validation (16) of the 30-item PSQ (0.35 ± 0.14, range 0.08–0.86) and the healthy adults in the German revision (0.33 ± 0.17, range 0.00–0.85) also do not differ.
Values are listed in Table 4, and differences are roughly summarized in Figure 2 (see Table 5 for details). All scales differed between patients and healthy adults. The extent of the differences varies with the scale and the group in question. The most severe stress values are obtained from Psychosomatic Medicine patients, especially outpatients, followed by tinnitus patients. Both groups have higher stress levels than IBD patients and women after spontaneous abortion, who report the second highest. Next in line are students, pregnant women, and women after regular delivery. Consistently low stress levels are reported by healthy adults. Students selectively report high levels of demands. Pregnant women and women after regular delivery show the highest levels of joy and the lowest levels of worries. When controlled for age, they even have significantly better values than healthy controls.
Sociodemographic differences were tested on the healthy adults sample (n = 334). All scales are significantly associated with age (worries r = −0.14*, tension r = −0.25**, joy r = 0.14**, demands r = −0.31**, overall score r = −0.28**), but not with gender. Figure 3 presents differences between age groups. Perceived stress is diminished and joy is raised in age groups over 60 years. Only worries are slightly lessened in the 60s group but no longer significantly in the 70s and older. Demands are selectively elevated in the 30- to 39-year-olds.
To test for immunological differences, women after miscarriage were divided in two stress groups by median-split. Decidual tissue for immunohistochemistry could be obtained in 50 cases. Women with a higher stress score had a significantly higher rate of tryptase+ mast cells (TMC+), of CD8+ T-cells, and of TNF-α+ cells (Figure 4). No differences resulted with regard to CD56+ NK-cells and CD3+ T-cells. In sum, higher perceived stress scores are associated with some of the relevant indicators of a supposed immunological imbalance in women who have had a miscarriage (22).
Sensitivity of Change
Psychosomatic Medicine inpatients under treatment show significant improvements for three of the stress scales and the overall score, but no change for joy. Tinnitus patients after 10 weeks of relaxation training (27) show a significant decrease of tension and an increase of joy, whereas worries and demands remain unchanged (Figure 5).
It took respondents on average 4.9 minutes to complete the revised 20-item questionnaire. The time median was 3.3 minutes. Only 5% of the patients took longer than 10 minutes; a few of those took up to an hour. We could not find any differences between diagnostic groups. There also was not any indication of a language effect in non-German-born participants. Age was significantly correlated with time-to-complete (r = 0.28). The stress scores themselves were slightly negatively (!) correlated with time (between r = −0.05 and −0.06), the less stressed patients taking more time to complete, yet when controlling for age, this association disappeared.
The PSQ by Levenstein et al. (14) was revised and tested for its dimensional structure on a large sample. We reduced the length of the questionnaire from 30 to 20 items and explored a meaningful and widely stable structure. The scales are balanced in the sense of comprising of the same number of items. Reliability values and construct validity are satisfactory.
Exploratory analyses were performed on one sample, confirmatory analyses on a second and separate sample. The original 7-factor solution was not replicated when the complete 30-item scale was analyzed. Instead, a 4-factor solution emerges. SEM analyses confirm this structure. Multisample analyses yield a sufficiently stable dimensional structure across different subgroups of patients and healthy adults. On the whole, the structure appears statistically robust and satisfactory as regards content. Consequently, in our opinion, the problem concerning the path weight between item 29 and the demands factor can be considered of minor importance and does not justify a modification. A trend toward comparably lower path weights—as can also be observed in items 01 and 10—might arise from positive item wordings. However, considering that mixed item wordings have advantages of their own, such as representing various facets of the latent construct or keeping subjects attentive, we do not endorse abandoning the positively worded items.
The dimensional structure is meaningful. Three factors (worries, tension, and joy) represent the dimension of stress reactions. In our opinion, the positively coded joy scale could assess a positive challenge or a personal resource component. The fourth factor (demands) represents a specific aspect of perceived environmental stressors. That the demands scale has a different focus than the three other scales is also proven by lower correlations of demands with the remaining scales. To regard the demands scale as focusing on an environmental dimension of perceived stress and the other scales as focusing on perceived stress reactions would be in line with findings from earlier studies in which the person's perception of stress was best represented by the two global dimensions of external stressor and stress reaction (2). Differential validity of the demands scale is supported by two findings: Students report higher levels and older adults report lower levels of demands.
Demands can be considered external stressors (8,21). However, the scale does not claim to cover all possible external stressors. Item topics are confined to the perception of basic demands on one's performance, like having too many things to do or being under time pressure. We do not know what specific demands a person who scores high on that scale has in mind. Specific hassles or life events are not included in this questionnaire.
Furthermore, an explicitly social component of environmental stressors is not included. For instance, out of the original “harassment scale,” which had dealt specifically with interpersonal tensions, only one of four items remained (“You feel that too many demands are being made on you”). The harassment scale had originally explained the greatest share of variance (15%) and it had strongly correlated with physical outcome. This is a possible limitation of the briefer PSQ. In sum, construct validity results point out that the psychological component of perceived stress is well represented by the 20-item PSQ, whereas the social component is not. Therefore, studies that strongly focus on social stress issues should prefer the use of the original 30-item questionnaire.
Future research could endeavor to strengthen and differentiate the stressor side of the questionnaire and to economize the stress reaction side with the aim to assess both sides of the coin accurately and economically. The relative merits of presenting a specific time frame, as in the original PSQ, or leaving it open-ended, as in the instructions for this revision, also remain to be assessed. As the time-frame depends on the specific research question, future research should specify and compare different time-frames.
Comparisons of PSQ index values between gastroenterological samples of the study by Levenstein et al. (14) and the present study yield no differences in measurement precision or respondents' scoring. This indicates that—concerning the index score—the revised German version of the questionnaire reaches the same precision as the English original, with comparable results. In the original study, Levenstein et al. (14) observed higher values for Italian than for English respondents. Considering our own results, we do not expect great deviations between German and English-speaking samples, yet we consider a confirmation of the revised questionnaire with an English-speaking sample desirable. Only if structural invariance between samples from different cultural or lingual backgrounds was substantiated could we confidently use the instrument for studies across cultures.
Similar to the original questionnaire, the revised instrument does not significantly vary with respect to gender. This is not consistent with other research, which yields higher perceived stress scores for women (17,20). An explanation for this could be that the original PSQ was specifically designed and developed to ensure that men and women would have similar scores (14).
As to age, the original study yielded a relatively small correlation between the overall score and age (r = 0.22). In the present study, the association with age is reversed (r = −0.28). This might be due to the fact that among the former sample, older age groups were underrepresented (mean age was 32), whereas the sample of the present study covers all age groups. Here, group differences suggest that the demands values are slightly higher for the 30- to 39-year-olds compared with the 20- to 29-year-olds. This would be in line with the former findings. However, the overall stress score is appreciably lowest for the age groups above 60 years. Those groups were hardly represented in the original study.
Reference values for healthy adults and different disease groups were attained. We found particularly high stress levels in Psychosomatic Medicine patients, followed by patients with tinnitus and IBD and women after spontaneous abortion. Women in pregnancy or after regular delivery and healthy adults report the lowest stress levels. The data prove differential validity. Decreased levels of perceived stress after different forms of treatment in different settings sufficiently substantiate sensitivity to change.
In sum, our revision of the PSQ arrived at an economic, reliable, structurally stable and valid instrument that enables us to assess perceived stress in healthy adults and different disease groups. It measures three dimensions of a stress reaction (worries, tension, joy/reversed) and one stressor dimension. Because the stressors are generic, the questionnaire can be administered to different clinical and healthy adult samples in different settings. Results can be compared with the reference values at hand and across studies. The overall score is comparable to results from earlier studies with the original instrument (14,16). The original 30-item questionnaire's structure was not replicable, whereas the 20-item version's structure proved reasonably robust. Taking this advantage and respondent burden into account, we suggest that the 20-item version is preferable. However, it means that notably the social stressor domain is not sufficiently represented. Furthermore, future research should also investigate how a corresponding 20-item English version of the PSQ would perform.
We wish to thank Ingrid Wittmann, Urania Berlin, and Jan Schwendowius for their assistance in raising the healthy adult sample, and especially Dr. Susan Levenstein for her many helpful comments on the paper.
1. Kenny DT, Carlson JG, McGuigan FJ, Sheppard JL. Stress and health: research and clinical applications. Amsterdam: Harwood Academic; 2000.
2. Lobel M, Dunkel-Schetter C. Conceptualizing stress to study effects on health: environmental, perceptual, and emotional components. Anxiety Res 1990;3:213–30.
3. Cohen S, Kessler RC, Gordon GL. Strategies for measuring stress. In: Cohen S, Kessler RC, Gordon LU, editors. Studies of psychiatric and physical disorders in measuring stress. New York: Oxford University Press; 1995. p. 3–26.
4. Adler CM, Hillhouse JJ. Stress, health, and immunity: a review of the literature. In: Miller T, editor. Theory and assessment of stressful life events. Madison: International Universities Press; 1996. p. 109–38.
5. Critelli J, Ee J. Stress and physical illness: development of an integrative model. In: Miller T, editor. Theory and assessment of stressful life events. Madison: International Universities Press; 1996. p. 139–59.
6. Searle A, Bennett P. Psychological factors and inflammatory bowel disease: a review of a decade of literature. Psychol Health Med 2001;6:121–35.
7. Dohrenwend BS, Dohrenwend BP. Socioenvironmental factors, stress, and psychopathology. Am J Comm Psychol 1981;9:128–59.
8. DeLongis A, Coyne JC, Dakof G, Folkman S, Lazarus RS. Relationship of daily hassles, uplifts, and major life events to health status. Health Psychol 1982;1:119–36.
9. Lazarus RS, Folkman S: Stress, appraisal, and coping. New York: Springer; 1984.
10. DeLongis A, Folkman S, Lazarus RS. The impact of daily stress on health and mood: psychological and social resources as mediators. J Pers Soc Psychol 1988;54:486–95.
11. Cohen S, Kamarck T, Mermelstein R. A global measure of perceived stress. J Health Soc Behav 1983;1983:385–96.
12. Derogatis LR, Coons HL: Self-report measures of stress. In: Goldberger L, Breznitz S, editors. Handbook of stress: theoretical and clinical aspects. New York: The Free Press; 1993. p. 200–33.
13. O′Keefe MK, Baum A. Conceptual and methodological issues in the study of chronic stress. Stress Med 1990;6:105–15.
14. Levenstein S, Prantera C, Varvo V, Scribano ML, Berto E, Luzi C, Andreoli A. Development of the Perceived Stress Questionnaire: a new tool for psychosomatic research. J Psychosom Res 1993;37:19–32.
15. Levenstein S, Prantera C, Varvo V, Scribano ML, Andreoli A, Luzi C, Arcà M, Berto E, Milite G, Marcheggiano A. Stress and exacerbation in ulcerative colitis: a prospective study of patients enrolled in remission. Am J Gastroenterol 2000;95:1213–220.
16. Sanz-Carrillo C, Garcia-Campayo J, Rubio A, Santed M, Montoro M. Validation of the Spanish version of the Perceived Stress Questionnaire (PSQ). J Psychosom Res 2002;52:167–72.
17. Bergdahl J, Bergdahl M. Perceived stress in adults: prevalence and association of depression, anxiety and medication in a Swedish population. Stress Health 2002;18:235–41.
18. Wachirawat W, Hanucharurnkul S, Suriyawongpaisal P, Boonyapisit S, Levenstein S, Jearanaisilavong J, Atisook K, Boontong T, Theerabutr C. Stress, but not Helicobacter pylori, is associated with peptic ulcer disease in a Thai population. J Med Assoc Thai 2003;86:672–85.
19. Abell N. The Index of Clinical Stress: a brief measure of subjective stress for practice and research. Soc Work Res Abstracts 1991;27:12–6.
20. Koh KB, Park JK, Kim CH, Cho S. Development of the Stress Response Inventory and its application in clinical practice. Psychosom Med 2001;63:668–78.
21. Schulz P, Schlotz W. Trierer Inventar zur Erfassung von chronischem Streβ (TICS). Diagnostica 1999;45:8–19.
22. Arck P, Rose M, Hertwig K, Hagen E, Hildebrandt M, Klapp BF. Stress and immune mediators in miscarriage. Human Reprod 2001;16:1505–11.
23. Fliege H, Rose M, Arck P, Levenstein S, Klapp BF. Validierung des “Perceived Stress Questionnaire” (PSQ) an einer deutschen Stichprobe. Diagnostica 2001;47:142–52.
24. Angermeyer MC, Kilian R, Matschinger H: WHOQOL-100 und WHOQOL-BREF. Göttingen: Hogrefe; 1999.
25. Breivik E, Olsson U. Adding variables to improve fit: the effect of model size on fit assessment in LISREL. In: Cudeck R, du Toit S, Sörbom D, editors. Structural equation modeling. Lincolnwood: Scientific Software International; 2001. p. 169–194.
26. Browne M, Cudeck R: Alternative ways of assessing model fit. In: Bollen K, Long J, editors. Testing structural equation models. Newbury Park: Sage; 1993. p. 136–162.
27. Weber C, Arck P, Mazurek B, Klapp BF. Impact of a relaxation training on psychometric and immunologic parameters in tinnitus sufferers. J Psychosom Res 2002;52:29–33.