1 Personality, psychosocial risks at work, and health Katharine Parkes, PhD Department of Experimental Psychology, University of Oxford, UK A report prepared for the Collège d expertise sur le suivi statistique des risques psychosociaux au travail and commissioned by the DREES March 2010
2 Summary This report responds to questions formulated by the Expert Committee about the role of personality in the process whereby work-related psychosocial risk factors are implicated in mental and physical health impairment. There are four main parts to the report. In Section 1, as a background to the specific questions raised, models of work stress are described, and the various pathways by which personality and work-related psychosocial factors jointly impact on health, are examined with reference to empirical findings. Section 2 documents a systematic review of journal articles (published ) describing prospective studies that evaluate work-related psychosocial risks and one or more personality measures as predictors of mental and physical health outcomes. A total of 33 studies which met these and other pre-determined criteria were identified. Findings from the studies are summarised and discussed, with particular reference to evidence of the additive, interactive, and mediator roles of personality in relations between psychosocial risk factors and health outcomes. Section 3 responds to a further question concerned with changes in personality across the life course. Normative age-related changes in mean levels of personality variables are described, and work-related factors associated with individual change are considered, noting evidence of reciprocal influence between work experiences and personality change. Psychometric issues relevant to the development and validation of personality measures are outlined in Section 4, and the psychometric properties of several widely-used measures are described in Section 5. Particular attention is given to evidence of the reliability and validity of the personality measures identified in Section 2 as significant predictors of health outcomes in the prospective studies reviewed. Examples of personality measures available in the published literature, and the scoring methods used, are shown in the Appendix.
3 Index 1. Introduction 1.1 Demand-control-support model Effort-reward imbalance model Organizational justice model The role of personality in work stress 2 2. Psychosocial risks, personality and health: A systematic review 2.1 Background Literature search Results Findings for individual personality characteristics Conclusions Personality change over the life course 3.1 Mean-level changes in personality across the life course Individual differences in patterns of personality change Implications Psychometric issues: Reliability and validity of personality tests 4.1 Scale development Reliability and validity Fairness in psychometric testing Psychometric properties of specific personality measures 5.1 The NEO Five-Factor personality measures Over-commitment Negative affectivity Hostility Sense of coherence Locus of control Core self evaluations General issues References 64 Appendix: Examples of selected tests 80
4 1 1. Introduction Evidence from prospective studies shows that exposure to work-related psychosocial risks has an adverse impact on long-term health. Much of this evidence comes from research based on two current models of work stress, the job demand-control-support model (DCS) 1 and the Effort-Reward model (ERI) 2. These main features of these models, together with a more recent model, the Organizational Justice model 3, are outlined below. 1.1 Demand-control-support model In the DCS model, high demands (e.g. time pressures, work overload), low control (few opportunities to make decisions at work, limited skill utilization), and low social support are predicted to lead to high psycho-physiological strain and, over time, to adverse health outcomes. The significance of the DCS dimensions for health outcomes, particularly cardiovascular disease 4-6 and affective well-being 7-9 has been widely demonstrated. Significant findings have also been reported for other outcome measures, including minor health complaints 10, workability 11, absenteeism 12, and suicide/attempted suicide 13,14. However, the relative importance of the three DCS dimensions in predicting health outcomes varies across studies. Moreover, concerns have been raised about methodological limitations, particularly in relation to cross-sectional survey studies 7,15. Relatively little evidence supports the demand x control interaction originally predicted by the job strain model 16 ; a review of high quality longitudinal studies concluded that there was only modest support for the interactive hypothesis 7. More usually, additive effects are reported 17, Effort-reward imbalance model The ERI model proposes that an imbalance between effort (e.g. extrinsic job demands, responsibilities, and obligations) and rewards (money, promotion prospects, job security) is a risk factor for poor health 2. Unlike the DCS model, recent versions of the ERI model include an intrinsic personality component, designated over-commitment (OC). The nature and hypothesised role of OC has evolved in the development of the ERI model; it now refers to a personality trait combining excessive striving with needs for approval and esteem, and is regarded as a potential moderator variable 19. Findings from prospective studies support the ERI model in that high effort coupled with low reward is associated with poor mental and physical (particularly cardiovascular) health 20,21. Job insecurity has been found to add to the adverse effects of ERI 22. However, a recent study questions the value of combining effort and reward into a single measure 23, and other evidence suggests that causal relations between ERI and health may be reciprocal rather than unidirectional 24. The role of over-commitment has been less widely examined, but it was found to be significant as an additive risk factor, over and above ERI measures, in four out of five studies of CVD incidence, and in five out of eleven studies of CVD symptoms in a review
5 2 of ERI research 19. However, the review provided little support for the interactive model although a more recent survey, using measures from both the ERI and the DCS models, reported that low control (DCS model) and high OC (ERI model) combined synergistically to give rise to a high levels of depressive symptoms Organizational justice model Lack of organizational justice in the workplace has been recognised as a psychosocial risk factor that can lead to adverse mental and physical health outcomes 3. The Organizational Justice model has two components, procedural injustice (decisions at work lack consistency, openness and input from all affected parties) and relational injustice (lack of considerate and fair treatment of employees by supervisors); these components have been found to predict sleeping problems 26, poor mental health 27 and cardiovascular mortality 28 in prospective studies and to explain variance in health outcomes over and above that accounted for by ERI measures The role of personality in work stress The models outlined above do not incorporate individual personality characteristics as predictors of mental and physical health outcomes (with the exception of the OC measure in the ERI model). However, personality is known to be significant in relation to longterm health 29-31, and several work stress models include paths among personality traits, objective and perceived work stressors and health outcomes. One such model, the Michigan model 32,33 has been particularly influential in guiding research into the joint effects of personality variables and psychosocial work stressors, and is used as the basis for the present discussion. As represented in this model, shown in Figure 1.1, objective work characteristics influence subjective perceptions of work stress; these perceptions give rise to short-term affective, cognitive, behavioural and physiological responses which, with continued stressor exposure, lead to chronic long-term health impairment. However, the model also incorporates bi-directional pathways and feedback loops; for instance, long-term health impairment may lead individuals to perceive their work conditions less favourably, or to seek an objectively less demanding job. The influence of individual differences operates at several points in the stress process represented by the Michigan model. For instance, personality traits (and other individual characteristics) may act by influencing selection into different types of job and hence exposure to objective stressors 40, or by influencing work perceptions 34, or by directly influencing stress responses and health 35,36. Moreover, the model includes not only direct effects of personality, but also mediating and moderating effects, and bi-directional paths. Personality variables, particularly negative affectivity, are potentially involved in each of these mechanisms 37.
6 MODERATOR VARIABLES: INDIVIDUAL AND SITUATIONAL CHARACTERISTICS Social support Personality Demographic Behavioural Genetic Management Coping resources e.g. age, gender, e.g. exercise e.g. family history style education diet of illness Objective work stressors Short-term responses Long-term outcomes Work overload Long work hours Paced work, time pressures Lack of control over work tasks Shift work Organizational re-structuring, down-sizing, job insecurity Perceived work stressors Affective Cognitive Behavioural Physiological Medical e.g. Cardiovascular disease Psychological e.g. Chronic depression Anxiety disorder Behavioural e.g. Alcoholism Figure 1.1 Conceptual model of the stress process Note: Solid lines represent direct effects among variables. Broken lines represent interaction effects. Adapted from Israel et al. (1996) 3
7 Personality and work stress: Mediating effects Mediation refers to an indirect process by which the effect of one variable on an outcome measure is transmitted through an intervening variable (see Baron and Cohen 38 for statistical methods of testing mediation effects). Thus, personality may influence how individuals perceive the objective work demands to which they are exposed and these perceptions in turn may lead to short-term affective, physiological, and behavioural responses, and eventually to chronic health effects. In this presumed sequence, the effects of personality on long-term health are mediated through perceived work stressors and short-term stress responses. In a longitudinal study of mediation effects using a cross-lagged panel design, Schwarzer et al. 39 examined the role of self-efficacy in the process by which work stress led to burnout in teachers. The results showed that self-efficacy was a protective factor; resourceful individuals high in self-efficacy experienced less job stress, which in turn reduced subsequent burnout. Moreover, over the one-year follow-up period, the path from earlier self-efficacy to later burnout was stronger than the (non-significant) reverse causal path. Similarly, a measure of personality resources assessed in childhood and early adulthood was found to predict job satisfaction in middle adulthood, and this relationship was partially mediated by job complexity 40. A different form of mediation was reported by Kivimaki et al. 41 from a 7 yr follow-up study of hostility and sickness absence in which the personality variable Sense of Coherence (SOC) 42 acted as mediator, reducing the association between individual hostility and subsequent sickness absence by 33-50% depending on the outcome measure. In a further study, Feldt et al. 43 found that the longitudinal relationship between psychosocial stressors (including job insecurity) and psychosomatic/affective outcomes was mediated by SOC. These two studies suggest that change in SOC may result from exposure to individual and work-related psychosocial factors Personality and work stress: Confounding effects As used in work stress research, the term confounding refers to the role of personality traits (or other factors such as demographic or socioeconomic variables) in creating an apparent link between measures of work-related psychosocial risks and health, which may be solely or partly due to a third factor effect, the influence of the confounding variable(s) on both perceptions of job characteristics and health. In relation to personality, attention has been focused primarily on neuroticism/na as a potential confounding factor, i.e. as a source of self-report bias that should be statistically controlled in studies that seek to link perceived job stressors and health outcomes 44,45. Accordingly, self-report studies of psychosocial stressors and strain responses often include NA as a control variable 46,47. If inclusion of NA in a multivariate analysis reduces the relationship between perceived stressors and outcome, then confounding is a possible explanation (although it should be noted that the statistical test for confounding is the same as that for mediation; interpretation depends on the nature of the variables concerned, and the underlying theoretical viewpoint).
8 5 The view that NA is simply a nuisance variable and a potential source of bias has been disputed by other researchers, particularly Spector and his colleagues 34,48, who argue that there is little evidence to support a general bias effect that cuts across all stressor and strain variables. They suggest that NA plays an important role in work stress processes which should be addressed rather than statistically controlled, and they describe a series of substantive mechanisms through which NA could affect job stressors and strains, each supported by research evidence 34 : Perception. This mechanism reflects the tendency of individuals high in NA to view the world in a negative light, but their negative self-reports of job stressors are seen as valid indicators of their actual perceptions and experiences. Hyper-responsivity. High NA individuals may respond to the same objective level of stressors more strongly than their low NA counterparts, in which case NA would act as a moderating variable in relations between job stressors and affective responses. Selection. High NA individuals may be in more stressful jobs than those low in NA, because they are recruited for less favourable jobs, or because of self-selection. Stressor creation. High NA individuals may, by their own behaviour, create job stressors for themselves, for example, by creating interpersonal conflicts at work, or by managing their workload less effectively than low NA individuals. Mood. Transitory mood may affect the assessment of NA. If so, and if mood is also affected by job conditions, then a correlation between NA and job stressors could reflect the indirect influence of job stressors on reports of NA, rather than the effects of NA on reports of job stressors. Causality. Exposure to high levels of job stressors may tend to make individuals higher in NA; thus, this mechanism proposes that job characteristics affect the trait level of NA, as assessed empirically. Whilst these possible mechanisms are discussed in relation to NA by Spector et al. 34, they are not necessarily specific to NA; for example, Hoge et al. 49 found that sense of coherence had not only direct effects on strain measures, but also showed significant perception, selection, and stressor-creation effects, although work stressors remained substantial predictors of strain Personality and work stress: Reciprocal effects and reverse causation Although theoretical models represent job stressors as having a causal effect on health outcomes, evidence of reciprocal and reverse causal relationships between work-related psychosocial risk factors and health outcomes is increasing. In some instances, also, personality variables are implicated in these reverse or reciprocal relationships. A review of longitudinal field studies of organizational stress, found that about half of the 43 studies evaluated reverse causation in stressor-strain relationships, and some evidence of reversed causal effects was found in about 33% of the studies concerned 50. More recently, in an evaluation of the demand-control model, Dalgard et al. 51 found evidence of reverse causation over an 11-yr follow-up period; in this case, it applied to job demands but not to job control.
9 6 Several studies have applied structural modelling to longitudinal panel data which allows evaluation not only of direct and reverse causation but also of reciprocal causal effects between exposures and health measures. Although at least one study using structural modelling has affirmed the theoretical causal ordering among DCS job characteristics and work-related psychological well-being 52, reciprocal longitudinal findings have also been reported 53. Similarly, in an evaluation of cross-lagged reciprocal relationships among job demands, work-home interference, and general health, a reciprocal model proved to be superior to both the direct causation and the reversed causation models 54. Structural modelling has also been used to evaluate causal paths between ERI measures and health; using three waves of data collection, Shimazu et al. 24 found cross-lagged and causally dominant effects of ERI on employee psychological and physical health, but also evidence of reciprocal effects. Similarly, Xanthopoulou et al. 55 found that the best-fit model was one in which not only were personal resources (self-esteem, self-efficacy and optimism) and job resources (e.g. autonomy, support) reciprocally related to work engagement, but there was also a reciprocal relationship between job and personal resources. Taken together, these longitudinal studies do not provide strong support for uni-directional reverse causal paths from strains to stressors although, consistent with the ERI and DCS theoretical models, there is evidence of causal influences from psychosocial stressors to health-related outcomes. However, it is also clear that reciprocal paths play an important role in stressor-strain relationships; moreover, these reciprocal pathways may involve personality variables in addition to measures of psychosocial risk factors and health. Thus, even though personality characteristics are conceptualised to be stable over time, as empirically assessed, they appear to be subject to influence by work conditions over relatively short time periods Personality and work stress: Interactive effects Additive (or main) effects imply that two or more predictor variables contribute independently to explaining variance in an outcome measure; in contrast, interactive (or moderating) effects imply that the magnitude and direction of the effect of one predictor (for example, a psychosocial risk factor) on a outcome measure depends on the level of a second predictor (for example, a personality characteristic). Two forms of interaction can be identified, vulnerability/resilience and person-environment fit. Vulnerability implies that a high level of a maladaptive personality trait if combined with a high level of a psychosocial risk factor will result in a disproportionately adverse outcome relative to their additive effects. Thus, for instance, high NA individuals were found to show significantly greater reactivity to high job demands than low NA subjects 56. Conversely, resilience or buffering implies that a high level of an adaptive personality resource protects individuals from adverse effects of exposure to psychosocial risk factors.
10 7 Person-environment fit. In cross-over interactions, neither high nor low levels of a personality factor are necessarily maladaptive; rather, adverse outcomes arise from incongruence or lack of fit between personality and the environment. For instance, perceived job control and an individual s locus of control may be congruent or incongruent, and have favorable or unfavorable effects, respectively, on outcomes 57. Although the DCS model does not include personality variables, evidence that personality characteristics may act as moderators of the effects of exposure to high strain (or high isostrain) conditions has been reported. For instance, a significant three-way interaction was found to predict affective well-being in cross-sectional and longitudinal data 58 ; demand and control combined interactively for externals whereas additive effects were found for internals. Other researchers have also found that personality characteristics moderate the effects of DCS dimensions on health outcomes 57,59, and have advocated further research to extend the demand/control model with personal characteristics 59. The points in the causal process at which personality may moderate work stress effects are not yet clearly established. In particular, moderating variables may influence either or both links in the paths by which objective stressors influence perceptions of work stress, and perceptions of work stress are related to psychological and physiological responses. In particular, little is known about how personality variables may moderate relations between objective and perceived work stressors, although significant dispositional moderators of relations between objective and perceived characteristics of laboratory tasks have been identified 60.
11 8 2. Psychosocial risks, personality and health: A systematic review 2.1 Background Over the past five years, prospective research into psychosocial risk factors in relation to physical and mental health outcomes has been the subject of several systematic reviews. In particular, detailed reviews of coronary heart disease (CHD) 61,62 and mental health outcomes, including depression 21,63,64, in relation to work-related psychosocial exposures have been published. In these reviews, two of which include meta-analyses 63,64, risk factors defined by two theoretical models of work stress, the demand-control-support model (DCS) 1 and the effort-reward model (ERI) 2 play a major role. A further review, which includes both prospective and cross-sectional findings, focuses on the role of work stress in relation to coronary risk factors (including hypertension, blood lipids, and metabolic syndrome) 65. More specifically, Van Vegchel et al 19 examine research evidence for the extrinsic (effort reward imbalance) and intrinsic (over-commitment) components of the ERI model in relation to health-related outcomes. In general, and with some significant reservations, these reviews conclude that components of DCS and ERI models are prospective risk factors for heart disease and poor mental health, particularly depression. Indeed, referring to their meta-analytic review of work-related psychosocial risk factors in relation to mental health outcomes, Stansfeld and Candy 63 conclude that it provides robust consistent evidence that (combinations of) high demands and low decision latitude and (combinations of) high efforts and low rewards are prospective risk factors for common mental disorders (p. 443). Other authors have reviewed the effects of exposure to psychosocial risks from a wider perspective, considering both individual and environmental factors In particular, in a systematic review of prospective epidemiological data, Kuper et al. 68 concluded that there was evidence for the roles of depression, lack of social support, and psychosocial work characteristics on the aetiology and prognosis of CHD, but that evidence for an effect of anxiety or Type A /hostility was less consistent. In discussing this research, Kuper et al. draw attention to the difficulty of determining the extent of bias in the reporting of psychosocial findings; they also note that bias may occur after publication as strongly positive results are more likely to be cited by other papers than weak or negative findings. In the context of the present review, it is relevant that the 71 studies of CHD (published up to 2001) summarized by Kuper et al. include almost no prospective research that examines work-related psychosocial risk factors for CHD and personality measures. Moreover, in two studies reviewed by Kuper et al. that do report both work exposures and personality characteristics, the latter are used only as control variables, the main interest being the role of psychosocial work exposures 69,70. Conversely, prospective studies that focus on individual personality traits as predictors of long-term health outcomes rarely include psychosocial work exposures. Adverse health outcomes, such as all-cause mortality, cardiovascular mortality, coronary heart disease, depression and other mental health problems, suicide and attempted suicide, have been
12 9 linked to aspects of personality in long-term prospective studies. Thus, a 2006 review of the literature on the development and course of physical illness highlights the adverse effects of negative affectivity and anger/hostility, and the positive role of optimism 29, while more recent studies identify extraversion, conscientiousness, and sense of coherence as significant predictors of favourable long-term mental and physical health outcomes 30,71,72. Possible mechanisms underlying the effects of personality on long-term health include shared genetic influences, health behaviours, and the influence of personality on appraisal, coping, and physiological reactivity 29 ; personality may also influence exposure to potential stressors (including those in the workplace) and the stressreducing resources (e.g. social support) available. One possible explanation of the relative lack of long-term prospective research that analyses both personality measures and psychosocial work exposures is that the job strain/iso-strain model 1, which has attracted much research attention, only includes work environment characteristics; it does not take into account the possible role of individual differences in personality and coping in responses to adverse work conditions. Proponents of the DCS model point to evidence suggesting that individual differences are unlikely to account for the association between job strain and CHD 6, although the need for further research into interactions between environmental stressors and personality characteristics is recognized. An important exception to the paucity of research combining work exposures with individual personality characteristics is a study based on data from the Whitehall II study of UK civil servants 5. The analysis included measures of hostility, Type A behaviour, competitiveness, negative affectivity, minor psychiatric disorder, and two coping patterns, together with measures of job control, in relation to newly reported CHD events over a 5.3 yr follow-up period. The results showed that when age-adjusted odds ratios for the effects of low job control on CHD outcomes were compared across sub-groups identified as having or not having each negative personality characteristic, differences were small and not consistently in one direction, nor consistently in the same direction for men and women. These results were not substantially changed by including other control variables, by use of different analyses, or when job demand and social support were used as exposure measures. The authors concluded that adverse effects of low job control could not be explained by confounding effects of negative personal characteristics, or by a generalized tendency for neurotic individuals to complain. They also ruled out possible mediating or moderating effects of negative personality factors, concluding that their findings seem to justify the relative disregard of personal factors and individual differences concerning low job control (p. 406). However, the authors noted some psychometric problems with their measures, and suggested that other personal attributes, e.g. locus of control, or use of different work stress models, might produce different results. Moreover, this study only considered CHD outcomes, and the authors emphasised that personal characteristics should certainly not be neglected in the broader field of job stress research.
13 10 In contrast to the job strain model, the Effort-Reward Imbalance (ERI) model incorporates an intrinsic component, a personality characteristic designated over-commitment (OC), which reflects excessive striving combined with a strong need for approval and esteem. Studies reviewed by Van Vegchel et al. 19 suggest that there is evidence for a significant additive role of OC, but there is little support for the model that predicts that OC and ERI act synergistically to predict health outcomes. The aim of the present review is to bring together findings from prospective research which examines personality factors and work-related psychosocial risks as joint predictors of mental and physical health outcomes. The review focuses on two main questions: First, is there evidence that personality variables contribute to explaining health outcomes over and above the effects of work-related psychosocial risk factors? In this context, not only direct relationships, but also possible confounding, mediating, and/or reverse or reciprocal effects among personality variables, psychosocial risk factors, and health outcomes are potentially relevant. Second, is there evidence that personality factors act as moderators of relations between work-related psychosocial risks and health outcomes? As described below, a systematic search of the literature published from inclusive was undertaken to address these questions. 2.2 Literature search The main literature search was carried out using the ISI Web of Science, and two components of the Scopus database Health Sciences (which includes Medline) and Social Sciences & Humanities (which covers journals in Psychology and Social Sciences). Additional searches were carried out using OvidSP. Further material was found by examining journal articles listed as having cited the articles located in these searches. The search terms identified prospective/longitudinal studies published in peer-reviewed journals in the years , in which measures of work-related psychosocial risks, including (but not restricted to) dimensions from the Demand-Control-Support (DCS) model and the Effort-Reward Imbalance (ERI) model, together with one or more personality measures, were used to predict health-related outcomes. Search terms for personality included specific personality variables in addition to the general terms personality, disposition*, and trait*. The main search terms used are shown in Table Criteria for selecting studies The many documents located in these searches were examined to identify studies that met pre-determined criteria, specified with reference to recent systematic reviews of psychosocial risk factors in relation to health outcomes 63,64,68.
14 11 Table 2.1 Main terms used in the literature search Personality Personality Disposition* Trait* Vulnerable / vulnerability Resilient / resilience Negative affect* Positive affect* Neuroticism Extraversion Optimism Locus of control (LOC) Self-efficacy Sense of coherence (SOC) Hostility Type A Over-commitment (OC) Coping resources NEO inventory NEO-FFI Psychosocial work environment (Work* OR job OR occupation*) AND (stress OR psychosocial) AND Demand* Control Discretion OR decision authority OR decision latitude OR autonomy Social support Iso-strain OR job strain Effort AND reward Organizational injustice Organizational justice Work hours OR time pressure Workload Job security OR job insecurity Re-structuring OR down-sizing OR relocation Outcomes Mortality Cardiovascular Heart disease Coronary Psychiatric Mental health Distress Anxiety Depression Mood Affect* Psychosomatic Somatic symptoms Well-being Job satisfaction Illness Sickness absence Sick leave Turnover Method Longitudinal* Prospective* Follow-up Search terms in different groups were joined by AND, and those within groups by OR The following criteria were applied in selecting the studies to be included: Longitudinal studies of working age adults which included one or more psychosocial work exposures AND one or more personality characteristics as predictors of health-related outcomes.
15 12 Outcomes were cardiovascular disease, psychiatric diagnoses, mental and physical health, psychosomatic complaints, insomnia, job-related affective responses, and job-related behavioural outcomes (e.g. sickness absence, turnover), identified by hospital or company records, assessed by formal diagnostic interview, or by referenced self-report scales. Clear descriptions of the measures used. A follow-up period of at least one year Control for initial level of dependent measure, and/or exclusion of baseline cases. A study size of at least 200 participants Sample located in Europe, North America, New Zealand, Australia, Japan or Russia English language articles published in peer-reviewed journals, Must report statistical results relating to personality variables (i.e. not simply state that they were used as controls) A total of 33 studies met the criteria and were included in the present review; the relatively small number of studies selected was primarily due to the triple requirement for prospective data in which both personality and psychosocial risk exposures were assessed. As others have noted 73, the majority of studies in this research area are either crosssectional and/or do not include measures of both personality and work characteristics. 2.3 Results Details of the 33 studies included in the review are summarized in Table 2.2 (pages 26-38). The follow-up durations generally ranged from 1-10 years, although two studies used childhood measures. Both men and women were included in almost all the studies, but separate results were not always presented; more usually, gender was treated as a control variable. The analysis methods included multiple regression and structural modelling (with continuous variables) and, more frequently, logistic regression models (based on groups with different levels of exposure). The psychosocial exposure measures and types of outcome variables used are outlined below. The personality variables, their main effects on outcome measures, and interactions between personality and psychosocial risks, are then examined in more detail Psychosocial work exposures Almost half the studies reviewed used measure from (or conceptually similar to) the DCS model or the ERI model to assess psychosocial work exposures. DCS model. Ten studies used measures of work-related psychosocial exposures derived directly from the DCS model, or based on similar constructs [6, 7, 8, 10, 12, 16, 19, 21, 32, 33] a. Most of these studies assessed job demand, control, and social support, although a The numbers in square brackets in the text identify the studies summarised in Table 2.2
16 13 in some cases only two of the three measures were used. If only one of the three DCS measures was used, the study was included in the other exposures group. ERI model. Five studies included in the review were based on the ERI model [1, 2, 4, 9, 31]. All of these studies assessed effort, reward and effort/reward imbalance, although proxy measures were used in some instances, e.g. . Each of these studies assessed the intrinsic effort component of the ERI model, over-commitment (OC), but only two studies [2, 31] tested the OC x ERI interaction. DCS and ERI models. Three studies [11, 13, 29] used both ERI and DCS measures. Other exposure measures. Non-theoretically based measures were used to assess work exposures in 15 studies; these studies were more heterogeneous than those based on the DCS or ERI models. Some focused specifically on a single exposure measure, assessed subjectively, e.g. job control  or objectively, e.g. unemployment ; others used objective and subjective assessments of the same work exposure . In some studies, specific objective work events were assessed in addition to a general perceived work stress measure [e.g. 14]. One study based on the Organizational Justice model  used measures of relational justice and procedural justice as exposure measures Outcome variables Overall, half the studies used only one outcome measure, although those that assessed jobrelated affective responses tended to use several measure, e.g. scales from the Maslach Burnout Inventory (MBI). Four different types of outcome variables were identified in the studies reviewed. General mental health outcomes  to . The mental health outcomes assessed in these studies included general measures of mental health/psychological distress, specific measures of depression and/or anxiety, self-reported health, and insomnia. In most cases, outcomes were assessed by standard self-report scales, but in two studies [6, 7] by interviews carried out either face-to-face or by telephone. Job-related affective well-being  to . Six studies in this group used scales from the Maslach Burnout Inventory (MBI) 74, sometimes coupled with more general measures (e.g. fatigue), for assessing outcome [16, 17, 18, 20, 22, 24]. Participants in these studies were primarily teachers, healthcare workers, and other human services personnel, although three studies [18, 20, 25] were based on random samples of the general population (screened to include only employed individuals) or trades unions. The majority of the studies used non-theoretical measures to assess work characteristics, but three studies used measures from the DCS model [16, 19, 21]. Behavioural outcomes  to . Of the five studies that used behavioural measures, four derived the outcome data from formal organizational records relating to sickness absence rates [27, 30]; health-care usage  and physician visits . Self-reported voluntary job turnover was used as the outcome in the remaining study . One study
17 14  included measures from both the DCS and the ERI models, and one used measures of organizational justice ; in the remaining three studies, non-theoretically based measures (down-sizing, stressful work events, and work challenge/hindrance stress) were used [26, 28, 30]. Cardiovascular disease outcomes.  to . The three studies in this group analysed longitudinal data from the Whitehall II prospective study of UK civil servants; in each case the mean follow-up period was approximately yrs. Of the three studies, one study  used proxy measures of effort, reward, and OC to test the ERI model, while two studies [32, 33] used measures from the DCS model of work stress Personality variables used in the studies The majority of the studies reviewed assessed only one or two personality characteristics (20 studies reported one personality measure, and 5 studies reported two measures). In all, there were 56 instances in which personality measures were analysed in relation to one or more outcome variables. The personality variables most frequently reported were the OC component of the ERI model and neuroticism/na, but three other personality measures (hostility/type A, locus of control/hardiness, and sense of coherence) were each used in at least five of the 33 studies, although not always assessed by the same scales. Of the remaining 11 measures, 7 were used in only one study. Information about the main and interactive effects of the personality variables is summarised in Table 2.2. The statistical data are taken from the fully adjusted models, although several studies noted the problem of possible over-adjustment, e.g. by including as control variables, health behaviours and/or lifestyle variables which may act as mediators of relations between personality/psychosocial factors, and health outcomes. Additive effects and, if reported, tests of the moderating effects of personality on relations between psychosocial work exposures and outcomes are shown, together with notes on confounding or mediating effects. 12 of the studies reviewed reported testing interactions between personality and work exposures but, in several cases, only limited details were provided. Findings for each of the main personality measures used are reported below. 2.4 Findings for individual personality characteristics Over-commitment (OC) The personality characteristic of over-commitment describes individual attitudes, behaviours and emotions reflecting excessive work-related striving combined with a need for approval and esteem. The measure forms an integral part of the Effort-Reward Imbalance (ERI) model 75. A recent psychometric evaluation demonstrated that the current version of the OC measure and the other ERI scales have stable psychometric properties 76.
18 15 Main effects. Eight of the studies included a measure of OC, together with the effort and reward measures from the ERI model; three of these studies also used DCS measures. None of the studies that used an OC measure used any other personality measures. OC showed significant main effects, additively with ERI measures, in all except one of the studies . Among men, high OC was consistently predictive of adverse health-related outcomes, including psychological caseness , self-reported depression, anxiety, and/or somatic symptoms [2, 4], insomnia , poor subjective health ratings , and incident CHD . Risk ratios (if reported) were generally in the moderate range, , except for CHD for which the HR value for OC was 1.26 (CI ). Among women, the effects of OC were less consistent; for instance, OC did not predict subjective health ratings among women , nor was it a predictor of anxiety or somatic symptoms, although it did predict depression in women . Findings from one study suggested that high OC was a risk factor for professionals, but not for manual workers . Entirely non-significant findings for the main effects of OC were reported in only one study. In this case, separate components of the ERI model (but not OC or effortreward imbalance) were significant predictors of an objective measure of health care usage . Thus, the strongest results for OC as a psychosocial risk factor were found for men in relation to a range of health-related outcomes, primarily of a psychological or psychosomatic nature. Interactive effects. Although the ERI model includes predicted interactions between OC and effort-reward imbalance, only two out of the eight studies that used the OC measure reported testing OC x ERI interactions. In both cases [2, 31], the results were nonsignificant. The present review therefore provides no evidence that OC acts as a moderator of ERI effects in relation to mental health outcomes or incident coronary heart disease, thus contributing further non-significant results to the generally inconsistent evidence reviewed by Van Vegchel et al. 19. However, the failure to report tests of interactions in six of the studies concerned weakens any wider conclusions that can be drawn. Differences between men and women in the extent to which OC acts as a risk factor for adverse health outcomes suggest that interactions between gender and OC might contribute significantly explaining outcome variance Neuroticism / negative affectivity (NA) NA is used here to refer to both neuroticism and negative affectivity, both of which reflect emotional vulnerability, pessimism, and a general disposition to react negatively to life and work stressors; individuals high in NA tend to be anxious, easily upset, often moody or depressed, and focused on negative aspects of self, other people and the world in general. In the studies reviewed, NA was the most frequently reported personality characteristic; it was assessed in 11 of the 33 studies, distributed across all the four types of outcome variables. Five of the studies based on the DCS model and six of the non-theoretically based studies included NA, although none of the ERI studies did so. In six of the studies that used a measure of NA, other aspects of personality were also assessed. NA is often
19 16 regarded as a confounding variable in the link between psychosocial work stressors and health outcomes, and several noted that it was included for that reason [3, 7, 8]. Main effects. In 9 out the 11 studies, including all the DCS studies, NA was a significant risk factor in multivariate predictive models, particularly in relation to mental health outcomes [3, 7, 8, 14], but also for job-related affective outcomes [16, 23], and for CHD [32, 33]. Risk ratios for NA in relation to mental health outcomes were reported for two studies; the values were 1.98 ( ) men / 1.55( ) women , and 3.59 ( ) men/women , representing moderate to strong associations. NA was also found to be a significant risk factor for sickness absence in both men and women , but it was not related to voluntary job turnover . In the two studies of CHD incidence [32, 33], NA was a main focus of interest rather than being included only as a potential confounder. Adjusted for demographic factors, those in the highest NA tertile (top one-third of scores) were at significant, albeit not large, risk for incident CHD events, hazard ratio=1.32 ( ); controlling for job strain, health behaviours, and other potential confounders in the multivariate model had little effect on this finding . However, the positive dispositional counterpart of NA, positive affectivity (PA) did not show significant effects, nor did the affect balance score (the difference between PA and NA scores). A further study of NA in relation to CHD , based on a similar Whitehall II dataset, used the Relative Index of Inequality (RII) as the risk index. The RII represents the hazard ratio for the extremes of the observed score distribution. In this study the RII value for NA was 1.64 ( ). NA and inflammatory biomarkers were found to be independent predictors of incident CHD; there was no evidence of mediation effects. Confounding effects of NA. Three studies of mental health outcomes [3, 7, 8] explicitly noted the role of NA as a potential confounder of relations between psychosocial factors and mental health outcomes and reported the significance of NA as a main effect, but only one study reported data showing the effects of controlling for NA. In this study , the effects of high job demand (highest tertile scores) remained significant when NA was included in the multivariate model, and the reduction in RR values was small for both men and women. Interactive effects of NA. Two very different studies [16, 27] included tests of interactions between NA and psychosocial risk factors in predicting outcomes; both reported significant findings. Elovainio et al. 77  examined the roles of NA and hostility as moderators of relations between measures of organizational justice and sickness absence rates. In addition to a significant main effect of NA in both men and women, NA moderated the effect of relational justice on sickness absence rates among men, but not among women. Thus, consistent with a personality vulnerability model, the combination of a stressful psychosocial context (low relational justice) and high NA among men led to particularly high rates of absence.
20 17 Houkes et al. 78  used LISREL structural modelling to evaluate additive and interactive effects of NA in a cross-lagged panel design. In the additive model, inclusion of a direct causal path from Time 1 NA to Time 2 emotional exhaustion significantly improved the fit of the synchronous model; however, reverse causal and reciprocal paths were not significant. Interactive effects were evaluated by sub-group analyses; a significant moderating effect of NA on relations between workload and exhaustion was found. The total effect of Time 1 workload on Time 2 emotional exhaustion was.06 in the low NA group and.27 in the high NA group. This analysis confirmed and extended previous findings 56, and illustrated one of the substantive mechanisms identified by Spector et al. 34. In summary, NA was found to act as a significant additive risk factor in almost all the studies in which it was included, but there was little evidence of it acting as a confounding variable. There was also no evidence of reversed causal effects from psychosocial stressors to NA measures, or of reciprocal effects, although these were tested in only one study . Significant interactive effects were found in two studies [16, 27] both of which were consistent with a vulnerability model of NA Hostility / Type A behaviour Main effects of hostility /Type A behaviour. Although both hostility and Type A behaviour pattern have been linked with the incidence of cardiovascular disease 79,80, in the studies reviewed, these personality characteristics were assessed in relation to mental health and behavioural outcomes. Measures of Type A behaviour [20, 28], and hostility [27, 30] were each reported in two studies; in addition, one study  assessed both these characteristics, and one study used teacher ratings of hostility at age 8 yrs to predict adult health outcomes in relation to employment status . A significant effect of hostility was found in both the studies that used sickness absence as an outcome measure. In one study, the risk of absence was higher by a factor of among individuals high in hostility than among others . Similarly, the second study reported significant regression coefficients for hostility as a predictor of sickness absence in both men (.15*) and women (.03**) . Hostility, but not Type A, predicted increased depression over a 3-yr follow-up in a large French cohort . Type A was also nonsignificant in relation to health care usage , and in an exploratory path analysis in which job demand and workload, and several personality measures, were used to predict fatigue and exhaustion . Confounding/mediating effects of hostility/type A behaviour. There was no evidence in these studies that hostility or Type A acted as confounding variables in relations between psychosocial factors and health outcomes, or that its effects on outcome measures were mediated by psychosocial factors. Interactive effect of hostility/type A behaviour pattern. Three studies tested hypothesised interactions between psychosocial risk factors and hostility [5, 27, 30]. In each case, the predicted interactions were significant; high hostility was a risk factor for individuals experiencing stressful conditions, specifically, unemployment , low job control during