Social Science & Medicine 53 (2001) 603614

Sense of coherence and school-related stress as predictors of subjective health complaints in early adolescence: interactive, indirect or direct relationships?
Torbjorn Torsheima, *, Leif Edvard Aaroeb, Bente Wolda
a b

Research Centre for Health Promotion, University of Bergen, Christiesgt.13, N-5015 Bergen, Norway Department of Psychosocial Sciences, University of Bergen, Christiesgt.12, N-5015, Bergen, Norway

Abstract The role of sense of coherence (SOC) on the relationship between adolescent school-related stress and subjective health complaints was tested with structural equation modelling. As part of the crossnational WHO-survey Health behaviour in school-aged children 1997/98 Norwegian representative samples of 1592 grade 6, 1534 grade 8, and 1605 grade 10 students completed measures on SOC, school-related stress and subjective health complaints. A test of nested structural models revealed that both stress-preventive (D w2 814. 86, p50:001), stress-moderating (D w2 11.74, p50:02) and main health-enhancing (D w2 1289.1, p50:001) eects of SOC were consistent with the data. A model including all these relationships tted the data well (CFI=0.91, RMSEA=0.04). Age-group comparisons revealed that the association between SOC and stress grew weaker with age (p50:05), whereas the direct association between SOC and health complaints grew stronger (p50:001). The main eect of SOC accounted for between 39% (11 year olds) and 54% (15 year olds) of the variance in subjective health complaints. Findings indicate that SOC may potentially be a salutogenic factor in adolescents adaptation to school-related stress, and that relationships between SOC and healthy adaptation, may be evident in younger age-groups than previously anticipated. # 2001 Elsevier Science Ltd. All rights reserved.
Keywords: Sense-of-coherence; Moderator; Stress; Adolescence; Health-complaints; Resilience

Introduction Subjective health complaints like headache, backache, and abdominal pains, are common in early adolescence (Aro, Paronen, & Aro, 1987; Garralda, 1996; Goodman & McGrath, 1991; King, Wold, Tudor-Smith, & Harel, 1996; Mikkelsson, Salminen, & Kautiainen, 1997). A series of studies have implicated school-related stress in the development and maintenance of such health complaints (Aro et al., 1987; Garralda, 1996; Hurrelmann, Engel, Holler, & Nordlohne, 1988; Ystgaard 1997). However, the nding that not all students develop complaints from school-related demands has directed
*Corresponding author. Tel.: +47 55 58 33 01; fax: +47 55 58 98 87. E-mail address: torbjoern.torsheim@psych.uib.no (T. Torsheim).

the attention to factors that moderate the perception of stress, and the adverse health impact of stress (e.g. Wagner & Compas, 1990; Ystgaard, 1997). In adults, one of the stress moderators that has generated considerable interest is the sense of coherence (SOC), a global orientation to view life situations as comprehensible, manageable and meaningful (Antonovsky, 1987). In the original theoretical formulation Antonovsky (1987) proposed that SOC may inuence stress and health in three ways: (1) SOC inuences whether a stimuli is appraised as stressor or not; (2) SOC inuences the extent to which a stressor leads to tension or not; and (3) SOC inuences the extent to which tension leads to adverse health consequences. While research on adults in part support these assumptions (for a review, see Antonovsky, 1993), the role of SOC in child and adolescent health is largely unexplored. With the view that school adaptation has an essential impact on a wide

0277-9536/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved. PII: S 0 2 7 7 - 9 5 3 6 ( 0 0 ) 0 0 3 7 0 - 1

604

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

range of social, psychological, and behavioural outcomes, empirical evidence on the stress moderating role of SOC during adolescence may oer particularly scope for development of prevention policies. The aim of the present paper is to examine the ways SOC and schoolrelated stress interact in relation to subjective health complaints during early adolescence. SOC and stress appraisal In the processes linking life situations to health, stress appraisal is the rst process that SOC may inuence. Stress research indicates that level of ambiguity and uncertainty are important dimensions in appraisals of life situations. Unpredictable or incomprehensible life situations are potent sources of stress (Lazarus & Folkman, 1984). As a global orientation to life, the sense of coherence (SOC), will inuence the degree to which people view life demands as chaotic and incomprehensible, or coherent and comprehensible. Through the condence that . . .the stimuli deriving from ones internal and external environments are structured, predictable and explicable. . . (Antonovsky, 1987, p. 19), individuals with a strong SOC will be less likely to perceive ambiguity in encounters with life demands. In keeping with the hypothesis that a high SOC may help to appraise demands as non-stressful, studies on adult populations have reported moderate inverse associations between measures of SOC and measures of perceived stress. A review of these studies (Antonovsky, 1993) showed that that the associations are generally stronger for perceived measures of stress than for measures of stressful life events, suggesting a role in appraisal processes, and not in the actual exposure to stressful events. SOC and stress moderation As a next step in the stress process, SOC has been suggested to inuence coping expectancies in encounters with stress (Antonovsky, 1987, p.19). According to the transactional model of stress (Lazarus & Folkman, 1984), coping expectancies develop from secondary appraisal processes, where people assess the means that are available to deal with the stressful condition. As a global orientation to life, individuals with a strong SOC will have a general condence that resources are available to meet the demands posed by stressful situations (Antonovsky, 1987, p.19). This condence increases the likelihood of positive coping expectancies. In related conceptual formulations (e.g. Bandura, 1986; Kobasa, 1979; Ursin, 1988) coping expectancies are assumed to moderate reactions to stress. In line with these models, Antonovsky (1987) proposes that a strong SOC may help to prevent stress from turning into

potentially harmful tension. From this perspective SOC acts as a classic moderator of life stress. Empirical studies on the stress-moderating role of SOC show mixed ndings. In a study of Finnish adult workers, Feldt (1997) found that the relationship between work demands and health complaints was stronger for workers with a low SOC, but in statistical terms the interaction was weak. In a similar vein, Vahtera and colleagues (1996) found that job demands from active jobs lead to sickness spells in workers with low SOC, but not in workers with a high SOC. In contrast, a number of other studies have failed to detect stress-distress moderation (e.g. Anson, Carmel, Levenson, Bonneh, & Maoz, 1993; Flannery & Flannery, 1990), leaving the issue of SOC as a moderator unresolved. SOC and stress-termination As a third mechanism, Antonovsky suggested that a high SOC may prevent stress-associated tension from developing into health problems. Stressing the point that SOC is not a particular coping style, Antonovsky (1987) proposed that individuals with a high SOC are more likely to select the coping strategy that is ecient for dealing with the stressor. High SOC individuals tend to use problem-focused strategies, they are exible in their choices of strategies, and they are skilled in using feedback to redirect coping attempts. As a consequence, individuals with a high SOC are, in general, more likely to remove the source of stress, and to terminate the associated tension. Over time, individuals with a strong SOC will experience shorter periods of harmful tension than individuals with a weak SOC, suggesting a main eect between level of SOC and health. In line with the tension-termination hypothesis, a high SOC has been strongly associated with measures of selfreported health and well-being, as well as low scores on markers of disease (for a review see Antonovsky, 1993). While these ndings are in line with the tensiontermination hypothesis, authors have suggested that the strong associations to some extent may reect methodological confounding between measures of SOC and measures of self-reported health (Geyer, 1997; Korotkov, 1993). SOC and the mechanisms of health complaints Through the inuence on stress appraisals, coping expectancies and coping behaviour, SOC may aect processes that are essential in the development and maintenance of subjective health complaints. Uncertainty is a potent stimuli for the stress response. Once initiated, the prole of the stress response is moderated by response-outcome expectancies and control beliefs (Ursin & Hytten, 1992). Low perceived control over

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

605

stressful conditions have been associated with a general tonic activation involving all biological response systems, including changes in neuro-endocrine, vegetative, neuro-muscular, central-nervous and immune system functioning (Ursin, 1997). Activation is sustained when coping eorts to remove the stressor are unsuccessful, or when no attempts are made to remove the stressor (i.e. helplessness). Sustained activation causes long term sensitisation of neural transmission (Antelman, Soares, & Gershon, 1997; Dubner & Ruda, 1992; Woolf & Thompson, 1991). Recent contributions view long-term sensitisation as a candidate mechanism for chronically elevated levels of health complaints (Ursin, 1997). In sum, these nding suggest that stressful appraisals, negative coping expectancies and unsuccessful coping behaviour are associated with physiological processes that may permanently lower the threshold for experiencing subjective health complaints. SOC in adolescence In view of the pervasive impact that is claimed for the adult SOC, surprisingly little is known about the role of SOC in the normal adaptation of general adolescent populations. In the original theoretical formulation, Antonovsky (1987) emphasises that SOC is a developmental construct that becomes crystallised at the age of 30, suggesting a more uctuating and less essential role for SOC in earlier age-groups: The adolescent, at the very best can only have gained a tentative strong SOC, which may be useful for short-range prediction about coping with stressors and health status. (Antonovsky, 1987, p. 107). The hypothesised limited role for SOC in adolescent health, has been paralleled by a limited research focus on these groups. Some authors view adolescent experiences as important for the development of SOC in adulthood (e.g. Lundberg, 1997; Cederblad, Dahlin, Hagnell, & Hansson, 1994) but few contributions have addressed the potential health impact SOC may have during adolescence. The few studies on child and adolescent SOC, have to a large degree focused on particular risk groups, such as adolescents experiencing evacuation stress (Antonovsky & Sagy, 1986), learning disabled children (Margalit & Efrati, 1996), and adolescents with chronic disease (Baker, 1998). Contrary to the idea of a uid and weak adolescent SOC, these studies indicate that a young SOC may contribute to stress and coping in much the same way as does the mature adult SOC. While the above studies oer preliminary evidence that SOC may help adolescent risk groups to cope with particular dicult life conditions, the health impact of SOC may be even more far-reaching for normative demands that every adolescents encounter during the course of normal development. School-related demands

are potent sources of stress in adolescent normal populations (Eme, Maisak, & Goodale, 1979; Greene, 1988; Henker, Whalen, & ONeil, 1995). Elevated levels of such stress is associated with psychological distress (Wagner & Compas, 1990; Ystgaard, 1997) and somatic complaints (Aro et al., 1987; Garralda, 1996; Hurrelmann et al., 1988). As schooling is mandatory in most countries, exposure to school demands is beyond the control of adolescents. Identication of resources that may help to prevent stressful appraisals, or moderate the adverse health impact of stress, may serve as an important rst step in developing preventive strategies.

The present study Antonovskys (1987) original contribution suggests that the adolescent SOC may aect level of health complaints indirectly by preventing school-related stress appraisals, interactively by moderating the impact of stress, and directly by reducing the likelihood of sustained activation. A potential shortcoming in previous research, is the failure to compare the relative importance of these mechanisms at given developmental stages in life. To gain more knowledge on the role of SOC in early adolescents adaptation to school-related stress, the present paper investigates each of these assumed relationships: 1. SOC is inversely related to appraisals of schoolrelated stress. 2. SOC moderates the relationship between schoolrelated stress on subjective health complaints. 3. SOC is inversely related to health complaints.

Method Sample and sampling procedure As part of the WHO survey Health behaviour in school-aged children 1997/98 a representative sample of 5026 Norwegian 11, 13 and 15 year olds took part in the study, representing a response rate of 78%. 1733 was from grade 6 (mean age 11.46), 1623 was from grade 8 (mean age 13.46), and 1670 was from grade 10 (mean age 15.48). The sample was obtained using a clustered sampling procedure with school-class as the sampling unit (Currie, 1998). Clustered samples may potentially underestimate measurement error due to non-independent observations. However, previous documentation (King et al., 1996, P. 215) suggests there to be essentially no such design eect for self-reported health measures.

606

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

Instruments The HBSC symptom checklist This is an 8-item scale on reported symptoms (headache, abdominal pain, backache, depressed mood, irritability, nervousness, sleeping diculties, dizziness). In the last 6 months: how often have you had the following? Symptoms are rated on a vepoint frequency scale: about every day; more than once a week; about every week; about every month; rarely or never. (Cronbachs alpha : .76; test-retest: 0.80) Perceived school-related stress The HBSC school-related stress subscale measured this (Samdal, Wold, & Torsheim, 1998b). The subscale comprises three items, see Appendix B (Cronbachs alpha: 0.75; test-retest 0.80). Sense of coherence Antonovskys (1987) Orientation to Life Questionnaire, short form (SOC-13), was adapted to t early adolescents (see Appendix A). A description of the adaptations are made in Torsheim and Wold (1998). The 13-items tap into three components of comprehensibility, manageability and meaningfulness (Cronbachs alpha: 0.85, test-retest: 0.78). Procedure The survey was carried out according to a standardised protocol (Currie, 1998). Prior to the distribution of material, teachers received instructions how to administer the survey. Questionnaires were distributed and lled out during a regular school-hour. Students were informed that participation was voluntary, and that their responses were anonymous. Statistical analysis The t of the models were assessed with Structural equation modeling (SEM) procedures using EQS for windows ver. 5.4 (Bentler, 1995). When a theoretically founded causal model exists, SEM oers several advantages compared to linear regression analysis. SEM oers more sophistication in detecting violation of the model assumptions, and in comparison of alternative causal models. SEM may be particularly advantageous for the kind of interaction that the moderator model implies. In multiple regression, measurement errors tend to be high in interaction terms (Busemeyer & Jones, 1983), thus increasing the probability of type II errors in the tests of interaction. In SEM the latent variables are measured without measurement error, which reduces the problem of type II error in tests of interaction. Several procedures for latent

variable modeling of interaction (i.e. moderation) eects have been developed during the last decade (e.g. Jaccard & Wan, 1995; Kenny & Judd, 1984; Ping, 1996). The present study implemented the two-step procedure developed by Ping (1996). Prior to model estimation, the observed variables were centered to ensure uncorrelated main and interaction terms. A product term between the average of the indicators of the schoolrelated stress factor and the average of the indicators of the latent sense of coherence factor was computed. Estimation of the latent school-related stress *SOC interaction factor was done in two steps. Firstly, the errors and the loadings for the latent stress and the latent SOC factor was estimated. In the second step, the estimated errors and loadings were used to compute xed error and loading for the measurement equation of the latent interaction factor, leaving only the structural model to be estimated. Under the assumptions of normality and unidimensional measures, the method provides estimated solutions that are essentially identical to more elaborate methods (for a practical example see Ping, 1998). Goodness-of-t criteria Several indices of goodness of t exist. In the present study, the comparative t index (CFI) and the root mean square error of approximation (RMSEA) were used as the goodness-of-t-criteria. By convention, a CFI above 0.90 and a RMSEA lower than 0.05 is taken to reect adequate model t. Comparisons of nested models were made by testing the signicance of the dierence in w2 . According to Baron and Kenny (1986), a moderator eect would be indicated by a signicant improvement in model t when adding the eect of interaction factor to the main eects model. Measurement models The latent factors of the structural models were based on the following congeneric measurement models: For the latent school-related stress factor, the three items of the school-related stress subscale were used as three observed indicators of school-related stress. For the adapted SOC-13 scale, the summed scores for each of the three subcomponents meaningfulness, manageability and comprehensibility were used as indicators of the latent SOC factor. Feldt and Rasku (1998) showed that the three subfactors may reect one underlying factor. For the latent SOC*stress interaction factor, the product of the average of the three stress indicators and the average of the three SOC indicators were used as the observed indicator, with xed loadings and error variances. For the subjective health complaints latent variable, the eight observed complaint variables were used as indicators.

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

607

Missing values Cases with more than 30% missing values were excluded from the analysis. For the rest, missing values were replaced with variable means. The nal sample thus consisted of 1592 11 year olds, 1534 13 year olds, and 1605 15 year olds.

Results Table 1 shows the correlation matrix and the means and standard deviations that was used for the estimation of the structural equations. Preliminary analysis of measurement model Unidimensional measurement models is a prerequisite for obtaining unbiased results. The t of the measurement model for SOC, school-related stress and subjective health complaints was tested simultaneously by allowing the three latent factors to correlate. Unidimensionality for each of the latent factors would be indicated by a high overall goodness of t. The correlated factors model obtained a CFI of 0.96 and a RMSEA=0.03, suggesting a satisfactory, yet not perfect degree of unidimensionality. Close scrutiny of the measurement model suggested a minor departure from perfect unidimensionality in the observed indicators of the latent health complaints factor. Modelling the symptoms as two correlated latent factors, rather than the proposed one-factor model, provided a small improvement in the overall measurement model (CFI=0.98, RMSEA=0.02) but did not alter the error terms and factor loadings for the observed indicators of SOC and stress. To check for potential bias in the proposed one-factor model of health complaints the structural models to be presented in the next sections were estimated for both the one-factor solution and the two-factor measurement model. The substantive ndings did not dier for these solutions, suggesting no essential bias for the one-factor model. For parsimony, the results to be presented in the next sections is for the proposed one-factor measurement model of health complaints only. Model selection Structural models were tested as a series of nested models, moving from the most restricted model to models with less restricted assumptions. The most restricted model in the present study, was a model were school-related stress predicts health complaints but where SOC has no eects. This model served as a null model to compare the t of the three stresshealth mechanisms that SOC was assumed to inuence. As shown in Table 2, the null model showed a poor t

to the data, indicated by a CFI of 0.79. In this model, school-related stress was moderately associated with health complaints (Standardised path coecients, 11 year olds 0.35; 13 year olds 0.38; 15 year olds 0.32; p5 0:001 for all paths). To assess the t of the three mechanism, each of the proposed relationships were added in a stepwise manner. Model 1 included indirect eects only, model 2 included indirect and direct eects, and model 3 included indirect, direct, and interactive eects on health complaints. Model 1 (M1) tested the assumption that SOC inuences level of stress, but SOC was assumed to have no interaction with stress and no main eects on health complaints. M1 produced a relatively poor t to the data (CFI=0.84), but the t was better than the null model (Ddf 3, Dw2 814:86, p50:001), suggesting that the modelled relationship between SOC and perceived school-related stress was consistent with the data. Model 2 added the assumption that SOC has a direct eect on health, but SOC was not assumed to interact with stress. This model tted the data well, indicated by a CFI of 0.91, and a RMSEA of 0.04. Model 2 tted the data better than model 1 (Ddf 3, Dw2 1289:1, p50:001), suggesting that the assumption of a main eect of SOC on health complaints was consistent with the data. Model 3 tested the additional assumption that SOC and school-related stress interact in predicting health complaints. A signicance test of the dierence in w2 , revealed a marginally better t for model 3 (df 3, Dw2 11:74, p50:02), suggesting that also the modelled interaction eects tted the data. Invariance across age The best-tting model, model 3, incorporating both indirect, direct, and interactive eects of SOC on health complaints, was chosen as baseline model for testing invariance of structural coecients across age groups. To test if the path coecients were invariant across age groups, constraints of equality were imposed on the structural equations. Four equality constraints were entered simultaneously: Equality across groups for (1) the path coecient between SOC and school-related stress; (2) the path coecient between SOC and subjective health complaints; (3) the path coecient between school-related stress and subjective health complaints; and (4) the path coecient between the SOC*stress interaction factor and subjective health complaints. The constrained baseline model had a poorer t to the data than the unconstrained model (Dw2 , 31.40, df 8, p50:001), suggesting that the assumption of invariant path coecients did not t the data. To detect which of the path coecients that varied across age groups, constraints were released in a sequential manner. The

608

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

Table 1 Correlation matrix with means and standard deviations for the observed indicators 11 year olds 1. Stress1 2. Stress2 3. Stress3 4. SOC_ME 5. SOC_MA 6. SOC_CO 7. Headache 8. Abdominal 9. Backache 10. Feeling low 11. Nervous 12. Irritable 13. Sleep dic. 14. Dizziness 15. STR:SOC 13 year olds 1. Stress1 2. Stress2 3. Stress3 4. SOC_ME 5. SOC_MA 6. SOC_CO 7. Headache 8. Abdominal 9. Backache 10. Feeling low 11. Nervous 12. Irritable 13. Sleep dic. 14. Dizziness 15.STR:SOC 15 year olds 1. Stress1 2. Stress2 3. Stress3 4. SOC_ME 5. SOC_MA 6. SOC_CO 7. Headache 8. Abdominal 9. Backache 10. Feeling low 11. Nervous 12. Irritable 13. Sleep dic. 14. Dizziness 15.STR:SOC 1. 1.00 1.44 1.52 0.21 0.25 0.29 0.10 0.12 0.07 0.14 0.10 0.14 0.12 0.13 0.14 1.00 0.40 0.49 0.22 0.23 0.19 0.12 0.11 0.14 0.12 0.13 0.14 0.12 0.14 0.04 1.00 0.36 0.46 0.14 0.17 0.15 0.11 0.08 0.06 0.10 0.11 0.09 0.11 0.12 0.05 2. 1.00 1.57 0.26 0.27 0.32 0.09 0.17 0.08 0.12 0.11 0.16 0.13 0.13 0.13 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. M SD 0.00 1.04 0.00 0.94 0.00 1.06 0.00 2.20 0.00 2.79 0.00 3.43 0.00 1.07 0.00 1.06 0.00 0.88 0.00 0.99 0.00 1.14 0.00 1.04 1.00 0.00 1.35 0.24 1.00 0.00 0.85 0.04 0.05 0.78 2.39 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.27 1.00 0.00 0.08 0.06 0.63 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.27 1.00 0.00 0.08 0.10 0.56 0.97 0.93 0.98 2.24 2.83 3.38 1.12 1.01 1.09 1.08 1.14 1.05 1.26 1.03 2.17 0.96 0.96 0.94 2.19 2.65 3.20 1.18 0.90 1.19 1.09 1.11 1.05 1.25 1.05 1.97 13. 14.

1.00 0.25 0.28 0.31 0.10 0.17 0.12 0.17 0.17 0.18 0.17 0.13 0.12

1.00 0.47 0.48 0.15 0.16 0.16 0.28 0.24 0.21 0.19 0.18 0.05

1.00 0.76 0.22 0.24 0.21 0.37 0.34 0.30 0.25 0.24 0.11

1.00 0.24 1.00 0.26 0.38 1.00 0.23 0.22 0.23 1.00 0.38 0.23 0.27 0.23 1.00 0.31 0.25 0.28 0.18 0.40 1.00 0.34 0.21 0.29 0.22 0.36 0.30 1.00 0.27 0.21 0.24 0.19 0.27 0.28 0.26 0.27 0.34 0.30 0.28 0.25 0.25 0.30 0.11 0.04 0.05 0.04 0.07 0.04 0.10

1.00 0.52 0.23 0.24 0.23 0.16 0.16 0.09 0.14 0.14 0.21 0.12 0.13 0.07

1.00 0.29 0.26 0.27 0.17 0.13 0.11 0.20 0.20 0.21 0.14 0.15 0.04

1.00 0.48 0.50 0.18 0.19 0.13 0.28 0.25 0.24 0.22 0.18 0.09

1.00 0.77 0.26 0.28 0.20 0.50 0.42 0.36 0.29 0.27 0.13

1.00 0.26 1.00 0.28 0.39 1.00 0.24 0.26 0.24 1.00 0.48 0.31 0.37 0.23 1.00 0.37 0.30 0.34 0.20 0.47 1.00 0.39 0.25 0.30 0.21 0.43 0.39 1.00 0.31 0.21 0.22 0.22 0.30 0.24 0.27 0.28 0.43 0.33 0.29 0.31 0.28 0.25 0.10 0.04 0.04 0.06 0.11 0.05 0.07

1.00 0.50 0.24 0.31 0.30 0.09 0.14 0.08 0.18 0.14 0.15 0.11 0.14 0.12

1.00 0.21 0.23 0.22 0.08 0.10 0.09 0.16 0.13 0.12 0.13 0.14 0.03

1.00 0.43 0.46 0.14 0.14 0.16 0.33 0.26 0.19 0.21 0.22 0.05

1.00 0.75 0.28 0.29 0.23 0.53 0.41 0.35 0.34 0.29 0.17

1.00 0.25 1.00 0.25 0.38 1.00 0.20 0.26 0.26 1.00 0.48 0.31 0.32 0.24 1.00 0.39 0.29 0.28 0.18 0.50 1.00 0.36 0.22 0.29 0.16 0.41 0.35 1.00 0.30 0.20 0.23 0.19 0.29 0.26 0.26 0.27 0.37 0.34 0.28 0.29 0.28 0.28 0.11 0.07 0.06 0.09 0.14 0.08 0.08

order of releasing constraints was determined by incremental t indices from the Lagrange-multiplier test. As shown in Table 3, the largest improvement in model t was obtained by releasing the constraint of equal path coecients between SOC and health complaints (Model 3b, Dw2 15:96, Ddf 2,

p50:001), suggesting that this path coecient varied across age groups. A further improvement in model t was obtained by releasing the constraint of equal path coecients between SOC and school-related stress (Model 3c, Dw2 7:36, Ddf 2, p50:05), which indicates that the relationship between SOC and

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614 Table 2 Model summary for the tested models, multisample analysis Model (M0) (M1) (M2) (M3) Null model Stress appraisal Health resistance Stress moderation df 273 270 267 264 w2 3985.40 3170.54 1881.44 1869.70 CFI 0.79 0.84 0.91 0.91 RMSEA 0.05 0.05 0.04 0.04 Comparison M1 vs M0 M2 vs M1 M3 vs M2 Ddf 3 3 3 Dw2 814.86 1289.1 11.74 P

609

50.001 50.001 50.02

Table 3 Testing constraints of equal path coecients across groupsa Hypothesis (M3a) (M3b) (M3c) (M3d) (M3e)
a

df 272 270 268 266 266

w2 1901.10 1885.14 1877.78 1874.13 1873.40

CFI 0.91 0.91 0.91 0.91 0.91

RMSEA 0.04 0.04 0.04 0.04 0.04

Comparison M3b vs M3a M3c vs M3b M3d vs M3c M3e vs M3c

df 2 2 2 2

Dw2 15.96 7.36 3.65 4.38

p 50.001 50.05 ns ns

Path coecients equal across groups SOC ! Health complaints6across groups SOC ! stress6across groups Stress moderation6across groups Stress ! Health Complaints6across groups

Note. ns nonsignicant at the 0.05 level.

Fig. 1. Structural path model for direct, indirect and interactive relationships for SOC and stress on health complaints, multisample analysis on 11 year olds, 13 year olds and 15 year olds. aPath coecient: upper number 11 year olds; middle number 13 year olds; lower number 15 year olds. bPath coecients when no eects of SOC is included (null model).

school-related stress was dierent across age. For the remaining two path coecients, release of constraints did not improve model t (models 3d and 3e), which is consistent with the assumption of invariance across age groups.

Path coecients and eect decomposition Fig. 1 shows the path coecients of model 3c. Sense of coherence (SOC) was moderately associated with school-related stress, with the strongest inverse

610

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

association among 11 year olds (standardised path coecients: 11 year olds 0.47; 13 year olds 0.41; 15 year olds 0.40). SOC was inversely related to health complaints, with stronger relationship with increasing age (standardised path coecients: 11 year olds 0.59; 13 year olds 0.67; 15 year olds 0.71; p50:001 for all age groups). When controlling for the eects of SOC, the relationship between school-related stress and health complaints was very weak (standardised path coecients: 11 year olds 0.08; 13 year olds 0.07; 15 year olds 0.07; p50:05 for all age groups). An eect decomposition was done to assess the relative contribution for the dierent relationships regressed on subjective health complaints. Across age, as shown in Table 4, the main eect of SOC accounted for an increasing share of variance in health complaints, from 39% in the group of 11 year olds up to 54% in the group of 15 year olds. The interactive eects of SOC and school-related stress accounted for between 0.5 and 1% of the variance in health complaints.

Discussion In brief, we found that both stress-mediated eects, stress-interactive eects, and direct eect of SOC on health complaints were consistent with the data. SOC and school-related stress The reason for investigating the role of SOC was twofold. Firstly, a large literature indicate that many students perceive school-demands to be stressful. Secondly, the variance in these perceptions predicts variance in psychological and physical distress. Factors that prevent stressful appraisals, or that moderate the impact of stressful appraisals may be potential targets for preventive action. One of the assumptions made by Antonovsky (1987), is that individuals with a high SOC will tend to appraise demands as predictable and comprehensible. Based on Antonovsky, we expected SOC to be inversely related to perceived school-related stress. Students with a high SOC would be more inclined to view school demands as comprehensible and predictable, and less threatening to

well-being. In line with these expectations moderate-tostrong inverse relationships were found in all three age groups. The subcomponent of comprehensibility has been suggested to be an important factor in stress appraisal of recurring demands (Antonovsky, 1987). The role of comprehensibility may be particularly relevant in relation to school related demands, as the ambiguities posed by schoolwork is closely related to a lack of comprehension. The expectations of comprehensible demands may provide a cognitive set that organises the appraisal of such demands as they occur. The relationship between SOC and school-related stress grew weaker with age. With the notion of an emerging, and more and more crystallised SOC, this pattern was somewhat surprising. One possible explanation is that academic demands increase over age, and become less susceptible to benign appraisals. In the context of qualication to higher education, the potential threatening aspects of school-demands may become more salient. Such contextual factors may inuence stress appraisals more strongly than the level of SOC. As such, the relative contribution of SOC in appraisals of school-related stress may become lower when contextual factors dominate. Several authors have argued that school may be regarded as a work setting for adolescents (Rudd & Walsh, 1993; Samdal, Nutbeam, Wold, Kannas, 1998a). The size of the path coecients between SOC and school-related stress, was strikingly similar to what has been found in adult studies for work-related demands (Larsson & Setterlind, 1990; Ryland & Greenfeld, 1991). If we assume that schoolwork and paid work share essential features, such as performance demands and time pressure, our ndings may reect that that the adolescent SOC operates in much the same way as does the adult SOC. While several authors have stressed the labile character of the adolescent SOC (e.g. Antonovsky, 1987; Lundberg, 1997), the parallel ndings in this study do not support a strong distinction between the function of the adolescent and the adult SOC in appraisals of work demands. Does SOC moderate the impact of school-related stress? In the original theoretical formulation SOC is assumed to moderate the impact of stress. Individuals

Table 4 Eect decomposition of associations with subjective health complaints, best-tting model (standardised solution) Direct eects of SOC 11 year olds 13 year olds 15 year olds
a

Interaction 0.0119 0.0079 0.0055

Direct eects of stress 0.0060 0.0046 0.0042

Disturbance 0.5883 0.4998 0.4529

Total variance 1.001a 1.001 1.001

0.3944 0.4886 0.5388

Total variance exceeds 1.0 due to rounding errors.

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

611

high on SOC would expect stressors to be manageable. According to cognitive activation theory (Ursin & Hytten, 1992) such expectations may prevent the initiation of harmful activation during stress. The model that included interaction eects tted the data better than the model including main eects only. For students with a high SOC, school-related stress was less strongly associated with health complaints. It is tempting to interpret this in support for a stressmoderating role for SOC, but several methodological factors speaks against such an interpretation. When judging the present interaction-eects one has to consider that the study employed a method that minimises measurement error in the interaction term. The interactive eects were still only marginal, accounting for between 0.5 and 1% of the true variance in health complaints. Notably, the main eect of SOC on health complaints was strong. As suggested in the seminal paper by Baron and Kenny (1986), interaction eects may be dicult to interpret when the assumed moderator has a strong main eect on the outcome. From this point of view, it seems inappropriate to interpret the interaction eect as moderation. In any case, taking into account that other studies have shown mixed ndings of moderation eects, it seems reasonable not to overstate the substantive signicance of the present interaction eects. The weak interaction eects found in our study are in line with the bulk of studies reporting weak or nonsignicant interaction eects between SOC, stress and health outcomes (Anson et al., 1990; Feldt, 1997; Vahtera et al., 1996). For studies of stress-moderation in general, weak or absent interaction eects, has been explained with reference to lack of statistical power, and amplication of measurement error (Busemeyer & Jones, 1983). However, in the case of SOC, the weak interaction may also reside in the generality of the SOC construct. In social support research, a key issue has been that global support may not be able to match specic stressors (Gore & Aseltine, 1995). Thus, global resources should not produce strong moderation of domain-specic stress. The stressor-resource matching argument could also apply to SOC. The point that sense of coherence is not a particular coping style, or behavioural trait (Antonovsky, 1987), suggests that SOC may moderate stress through domain-specic response outcome beliefs, but these response outcome beliefs may be aected by a host of other situational factors. The moderating eect of SOC on stress, may thus be levelled out in specic situations. While global expectancies of being able to manage stress, may be relevant in encounters with school-related stressors, school-specic ecacy and skills may be even more instrumental in shaping coping expectancies. In future studies, instead of trying to establish moderation eects of SOC on domain-specic stressors, it may well prove

to be more fruitful to establish main eects between SOC and domain-specic self-ecacy and outcome beliefs. SOC and subjective health complaints The eect decomposition revealed that most of the variance accounted for by SOC could be attributed to the direct relationship between SOC and health complaints. In fact, the direct eects of SOC accounted for almost half of the variance in health complaints. The indirect and interactive relationships only accounted for a marginal proportion of variance. The strong main eect of SOC is open to several interpretations. According to Antonovsky, people with a high SOC use more eective coping strategies, and will always be moving from a state of tension to non-tension. In a more speculative vein, Antonovsky (1987) proposes that SOC aects the tendency for the physiological system to enter disequilibrium. Interestingly, this hypothesis converge with recent contributions that view subjective health complaints as a physiological sensitisation phenomenon (Ursin, 1997). The strong association between SOC and health complaints, controlling for level of stress, could reect that individuals with a high SOC show a resistance to such sensitisation. The strong associations do however bring forward the issue of a conceptual or methodological confound between SOC and subjective health complaints. Several authors have argued that measures of SOC in part reect negative aectivity or negative mood dispositions (Geyer, 1997; Korotkov, 1993). According to symptom perception theory, negative aectivity is associated with high introspection and low threshold for symptom perception (Watson & Pennebaker, 1989). Consequently, associations between SOC and health complaints could reect common inuence of negative aectivity. However, as argued by Stru mpfer and colleagues (1998), the association between SOC and negative aectivity may not primarily be an issue of methodological confounding. If low negative aectivity is viewed as emotional stability, the strong associations may well been seen as a validation of the stress-resistance component in SOC (Stru mpfer et al., 1998). While a high SOC may be a part of a larger adaptation pattern that includes low negative aectivity, SOC is not necessarily the same as negative aectivity (Stru mpfer et al., 1998). One possible interpretation is that SOC represents the cognitive-motivational manifestation of depression, and not the aective component per se. Cognitive models on depression (e.g. Beck, 1974) indicate that depressive beliefs causes aective states. The strong association with health complaints could mean that a low SOC reects the depressed belief-system

612

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

that sustain a high arousal, which must be clearly distinguished from the eects of such arousal. Alternative models The present paper investigates causal models for the relationship between stress, SOC and subjective health complaints, but the crossectional design does not allow for any rm conclusions regarding causality. This limitation needs to be particularly addressed when using structural equation modelling. While the results are consistent with the general mechanisms proposed by Antonovsky, several alternative models could t the data equally well. As a model worth considering, developmental perspectives suggest that SOC may mediate school-related stress. Load experiences from social role performance, have been assumed to have a strong impact on the development of SOC (Cederblad et al., 1994; Antonovsky, 1991). According to this view, experiences of persistent overload or underload, is associated with the development of a low SOC, whereas experiences of balanced load, promote a strong SOC. Following this argument, the adolescent SOC could be seen as inuenced by school-related stress, rather than as predicting school-related stress. Though conceptually very dierent, the SOC-asoutcome model would t the present data equally well as the models that were tested in our study. The problem of interpretation this leaves us with does not change the fact that there is a robust covariance structure between SOC, school-related stress and subjective health complaints. Future prospective studies could benet from adopting a transactional perspective on the relationship between SOC and stress, where reciprocal eects can be modelled.

rst step in providing a conceptual framework for how salutogenic factors may intervene on adolescent adaptation to school-related demands. Acknowledgements The authors would like to thank Candace Currie for preparation of the research protocol of the Health Behaviour in School-Aged Children 1997/98 survey. The writing of the manuscript was made possible through a doctoral grant from the Norwegian Research Council, division of medicine and health. Appendix A. The age-adapted SOC-13 Here is a series of questions relating to various aspects of our lives. Each question has ve possible answers. For each question, please mark the answer which best expresses your feelings about your life. 1. (Me) How often do you have the feeling that you dont really care about what goes on around you?1 2. (C )How often has it happened in the past that you were surprised by the behavior of people who you thought you knew well? 3. (Ma) How often has it happened that people whom you counted on disappointed you? 4. (Me) How do you think you are going to feel about the things you will do in the future? 5. (Ma)How often do you have the feeling that you are being treated unfairly? 6. (C) How often do you have the feeling that you are in a unfamiliar situation and dont know what to do? 7. (Me) How do you feel about the things you do every day? 8. (C) How often does it happen that you dont quite understand your own feelings and ideas? 9. (C) How often does it happen that you have feelings inside that you would rather not feel? 10. (Ma) Many people-even those with a strong character- sometimes feel like losers in certain situations. How often have you felt this way in the past? 11. (C) How often does it happen that you have the feeling that you dont know exactly whats about to happen? 12. (Me) How often do you have the feeling that there is little meaning in the things you do in your daily life?
1 Response keys for all questions except no.4 and no. 7 were: Very often } Often } Sometimes } Seldom } Never. For no.4 and no.7 response keys were : Like it a lot } Like it } Its OK } Dont like it } Dont like it all.

Conclusion The aim of the present paper was to assess the role of SOC in adolescents health adaptation to school-related stress. The results provide some support for the general stress-health mechanisms that Antonovsky formulated. While the present study did not address stability of SOC, the strong cross-sectional consistency across samples, indirectly point to a degree of stability in SOC also in adolescence. As such the present study underscores the need to adopt a life-span perspective when examining the role of sense of coherence in health. Importantly, the present ndings were established in relation to a kind of stress and health complaints that the general population of adolescents are exposed to during socialisation into adulthood. The cross-sectional design does not permit us to draw strong practical implications from the study, but the study does take a

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614

613

13. (Ma)How often do you have feelings that youre not sure you can keep under control.

Appendix B. The HBSC school-related stress subscale Do you agree or disagree with the following statements? (strongly agree } agree } neither nor } disagree } strongly disagree) V1. I have too much schoolwork. V2. I nd schoolwork dicult. V3. I nd schoolwork tiring.

References
Anson, O., Carmel, S., Levenson, A., Bonneh, D., & Maoz, B. (1993). Coping with recent life events: The interplay of personal and collective resources. Behavioral Medicine, 18, 159166. Antelman, S. M., Soares, J. C., & Gershon, S. (1997). Timedependent sensitization-possible implications for clinical psychopharmacology. Behavioural Pharmacology, 8, 505 514. Antonovsky, A. (1987). Unraveling the mystery of health. How people manage stress and stay well. San Fransisco, CA: Jossey-Bass Publ. Antonovsky, A. (1991). The structural sources of salutogenic strengths. In C. L. Cooper, & R. Payne (Eds.), Personality and stress: Individual dierences in the stress process (pp. 67 104). London: Wiley. Antonovsky, A. (1993). The structure and properties of the sense of coherence scale. Social Science and Medicine, 36, 725733. Antonovsky, H., & Sagy, S. (1986). The development of a sense of coherence and its impact on responses to stress situations. The Journal of Social Psychology, 126, 213225. Aro, H., Paronen, O., & Aro, S. (1987). Psychosomatic symptoms among 1416 year old Finnish adolescents. Social Psychiatry, 22, 171176. Baker, L. K. (1998). Sense of coherence in adolescents with cystic brosis. In H. I. McCubbin, E. A. Thompson, A. I. Thompson, & J. I. Fromer (Eds.), Stress, coping, and health in families. Sense of coherence and resiliency (pp. 107124). Thousand Oaks: Sage. Bandura, A. (1986). Social foundations of thought and action. New Jersey: Prentice Hall. Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological reasearch: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 11731182. Beck, A. T. (1974). Cognition, aect, and psychopathology. In H. London, & R. E. Nisbett (Eds.), Thought and feeling (pp. 127140). Chicago, IL: Aldine Publishing Company. Bentler, P. M. (1995). EQS structural equations program manual. Encino CA: Multivariate Software Inc.. Busemeyer, J. R., & Jones, L. E. (1983). Analysis of multiplicative combination rules when the causal variables are measured with error. Psychological Bulletin, 93, 549562.

Cederblad, M., Dahlin, L., Hagnell, O., & Hansson, K. (1994). Salutogenic childhood factors reported by middle-aged individuals: Follow-up of the children experiencing three or more childhood psychiatric risk factors. European Archives of Clinical Neuroscience, 244, 111. Currie, C. (1998). Health behaviour in school-aged children. Research protocol for the 1997/98 Survey. University of Edinburgh. Dubner, R., & Ruda, M. A. (1992). Activity dependent neuronal plasticity following tissue injury and inammation. Trends in Neurosciences, 15, 96103. Eme, R., Maisak, R., & Goodale, W. (1979). Seriousness of adolescent problems. Adolescence, 14, 9399. Feldt, T. (1997). The role of sense of coherence in well-being at work: Analysis of main and moderator eects. Work and Stress, 11, 134147. Feldt, T., & Rasku, A. (1998). The structure of Antonovskys orientation to life questionnaire. Personality and Individual Dierences, 25, 505516. Flannery, R. B., & Flannery, G. J. (1990). Sense of coherence, life stress, and psychological distress: a prospective methodological study. Journal of Clinical Psychology, 46, 415420. Garralda, M. E. (1996). Somatisation in children. Journal of Child Psychology and Psychiatry, 37, 1333. Geyer, S. (1997). Some conceptual considerations on the sense of coherence. Social Science and Medicine, 44, 17711779. Goodman, J. E., & McGrath, P. (1991). The epidemiology of pain in children and adolescents: A review. Pain, 46, 247264. Gore, S., & Aseltine, R. H. (1995). Protective processes in adolescence: Matching stressors with social resources. American Journal of Community Psychology, 23, 301327. Greene, A. L. (1988). Early adolescents perception of stress. Journal of Early Adolescence, 8, 391403. Henker, B., Whalen, C. K., & ONeil, R. (1995). Worldly and workaday worries: Contemporary concerns of children and young adolescents. Journal of Abnormal Child Psychology, 23, 685702. Hurrelmann, K., Engel, U., Holler, B., & Nordlohne, E. (1988). Failure in school, family conicts, and psychosomatic disorders in adolescence. Journal of Adolescence, 11, 237 249. Jaccard, J., & Wan, C. K. (1995). Measurement error in the analysis of interaction eects between continuous predictors using multiple regression: Multiple indicator and structural equation approaches. Psychological Bulletin, 117, 348357. Kenny, D. A., & Judd, C. M. (1984). Estimating the non-linear and interactive eects of latent variables. Psychological Bulletin, 96, 201210. King, A., Wold, B., Tudor-Smith, C., Harel, Y. (1996). The health of youth. A cross-national survey. WHO Regional Publications, 69. Kobasa, S. C. (1979). Stressful life events, personality, and health: An inquiry into hardiness. Journal of Personality and Social psychology, 37, 111. Korotkov, D. L. (1993). An assessment of the (short-form) sense of coherence personality measure: Issues of validity and well-being. Personality and Individual Dierences, 14, 575583. Larsson, G., & Setterlind, S. (1990). Work load/work control and health: Moderating eects of heredity, self image and

614

T. Torsheim et al. / Social Science & Medicine 53 (2001) 603614 Stru mpfer, D. J. W., Gouws, J. F., & Viviers, M. R. (1998). Antonovskys sense of coherence scale related to negative and positive aectivity. European Journal of Personality, 12, 457480. Torsheim, T., & Wold, B. (1998). Adolescent coping and health: Sense of coherence. In C. Currie (Ed.), Health behaviour in school-aged children. Research protocol for the 199798 Survey (pp. 7983). Edinburgh: University of Edinburgh. Ursin, H. (1988). Expectancy and activation: An attempt to systematize stress theory. In I. Florin, & H. Weiner (Eds.), Neurobiological Approaches to Human Disease (pp. 313 334). Toronto: Hans Huber. Ursin, H., & Hytten, K. (1992). Outcome expectancies and psychosomatic consequences. In B. N. Carpenter (Ed.), Personal coping. Theory, Research and Application (pp. 171 184). Westport: Praeger publishers. Ursin, H. (1997). Sensitization, somatization, and subjective health complaints. International Journal of Behavioral Medicine, 4, 105116. Vahtera, J., Pennti, J., & Uutela, A. (1996). The eect of objective job demands on registered sickness absence spells; do personal, social and job-related resources act as moderators?. Work and Stress, 10, 286308. Wagner, B., & Compas, B. E. (1990). Gender, instrumentality, and expressivity: Moderators of the relation between stress and psychological symptoms during adolescence. American Journal of Community Psychology, 18, 383406. Watson, D., & Pennebaker, J. W. (1989). Health complaints, stress and distress: Exploring the central role of negative aectivity. Psychological Review, 96, 234254. Woolf, C. J., & Thompson, W. J. (1991). The induction and maintenance of central sensitisation is dependent on Nmethyl-D-aspartatic acid receptor activation: implications for the treatment of post-injury pain hypersensitivity states. Pain, 44, 293299. Ystgaard, M. (1997). Life stress, social support and psychological distress in late adolescence. Social Psychiatry and Epidemiology, 32, 277283.

health behavior. International Journal of Health Sciences, 1, 7988. Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal and coping. New York: Springer Publishing Company. Lundberg, O. (1997). Childhood conditions, sense of coherence, social class and adult ill health: Exploring their theoretical and empirical relations. Social Science and Medicine, 44, 821831. Margalit, M., & Efrati, M. (1996). Loneliness, coherence and companionship among children with learning disorders. Educational Psychology, 16, 6979. Mikkelsson, M., Salminen, J., & Kautiainen, H. (1997). Nonspecic muskuloskeletal pain in pre-adolescents. Prevalence and 1-year persistence. Pain, 73, 2935. Ping, R. A. (1996). Latent variable interaction and quadratic eect estimation. Psychological Bulletin, 119, 166175. Ping, R. A. (1998). EQS and LISREL examples using survey data. In R. E. Schumacker, & G. A. Marcoulides (Eds.), Interaction and nonlinear eects in structural equation modeling (pp. 43100). London: Lawrence Erlbaum Associates. Rudd, R. E., & Walsh, D. C. (1993). Schools as healthful environments: Prerequisite to comprehensive school health? Preventive Medicine, 22, 499506. Ryland, E., & Greenfeld, S. (1991). Work stress and well being: An investigation of Antonovskys Sense of Coherence model. Journal of Social Behavior and Personality, 6, 3954. Samdal, O., Nutbeam, D., Wold, B., & Kannas, L. (1998a). Achieving health and educational goals through schools: A study of the importance of school climate and students satisfaction with school. Health Education Research, 13, 383397. Samdal, O., Wold, B., & Torsheim, T. (1998b). The relationship between students perception of school and their reported health and quality of life. In C. Currie (Ed.), Health behaviour in school-aged children. Research protocol for the 199798 survey (pp. 5155). Edinburgh: University of Edinburgh.