The Common Sense Model of Self-Regulation: Meta-Analysis and Test of a Process Model

According to the common-sense model of self-regulation, individuals form lay representations of illnesses that guide coping procedures to manage illness threat. We meta-analyzed studies adopting the model to (a) examine the intercorrelations among illness representation dimensions, coping strategies, and illness outcomes; (b) test the sufficiency of a process model in which relations between illness representations and outcomes were mediated by coping strategies; and (c) test effects of moderators on model relations. Studies adopting the common-sense model in chronic illness (k = 254) were subjected to random-effects meta-analysis. The pattern of zero-order corrected correlations among illness representation dimensions (identity, consequences, timeline, perceived control, illness coherence, emotional representations), coping strategies (avoidance, cognitive reappraisal, emotion venting, problem-focused generic, problem-focused specific, seeking social support), and illness outcomes (disease state, distress, well-being, physical, role, and social functioning) was consistent with previous analyses. Meta-analytic path analyses supported a process model that included direct effects of illness representations on outcomes and indirect effects mediated by coping. Emotional representations and perceived control were consistently related to illness-related and functional outcomes via, respectively, lower and greater employment of coping strategies to deal with symptoms or manage treatment. Representations signaling threat (consequences, identity) had specific positive and negative indirect effects on outcomes through problem- and emotion-focused coping strategies. There was little evidence of moderation of model effects by study design, illness type and context, and study quality. A revised process model is proposed to guide future research which includes effects of moderators, individual differences, and beliefs about coping and treatment.

The model also identifies emotional representations that reflect individuals' affective 23 responses to illness and may independently affect selection of coping procedures in parallel 24 to cognitive representations ( Figure 1). According to Martin, Leventhal, and Leventhal (2003), "affective and cognitive responses to symptoms may or may not be compatible. For 1 example, a person might understand that a symptom (e.g., bloody stools) could have serious 2 consequences if left untreated. This belief might motivate care seeking in the interests of 3 early detection. However, for some people, the fear of potentially receiving an ominous 4 diagnosis (e.g., of cancer) actually might inhibit self-referral behavior" (p. 210). Emotional  Illness representations are conceptualized as memory structures and may be activated 11 by a novel somatic change or salient information and lead to a search for a matching illness 12 prototype stored in memory. This is the first stage in the process of recognizing that one is ill, 13 and initiates self-evaluation (e.g., recognizing other symptoms, observing whether they go 14 away), self-management of illness (e.g., attempting to control the symptoms), and consulting 15 a professional (Leventhal et al., 2011). These stimuli are depicted to the left of Figure 1. The 16 progression to professional treatment may sometimes bypass both self-evaluation and self-17 management, when, for example, an acute event provokes immediate hospitalization leading 18 to diagnosis of illness, or when screening procedures identify diseases such as breast cancer 19 without any prior symptomatic information indicating to the individual that they are ill. 20 Also important are the moderating effects of socio-cultural and personal variables, 21 labelled the 'self-system' by Leventhal et al. (1992), that may influence interpretation of, and 22 response to illness stimuli, depicted at the top of Figure 1. For example, an individual may 23 perceive a headache and muscle ache as symptoms of a benign illness such as a common 24 cold. However, if such symptoms occurred after visiting a region where malaria is prevalent, they might activate representations of serious infectious disease associated with elevated 1 threat. Illness schema can be activated from sources which may be directly symptomatic and 2 concrete, but also arise independent of symptoms based on more abstract internal states that 3 interact with contextual knowledge (Orbell, Henderson, & Hagger, 2015). 4 Following the identification of illness and the activation of an illness schema 5 associated with an illness label (e.g., multiple sclerosis, diabetes, rheumatoid arthritis, cancer, 6 asthma), a dynamic self-regulatory process is initiated, comprising attempts to manage the 7 health threat and concomitant emotional reactions (Leventhal, Brissette, & Leventhal, 2003;8 Leventhal et al., 1980). These self-regulatory efforts are labeled coping procedures in the 9 common-sense model and follow from cognitive and emotional representations. For example, 10 an individual with bowel cancer may view the illness as having painful symptoms, a chronic 11 timeline, and controllable through dietary change, and therefore view a change of diet as an 12 effective means to cope with the disease progression. An individual who believes the same 13 disease is uncontrollable and finds it a source of emotional distress may adopt a denial coping 14 response. The parallel arrowed pathways in Figure 1 depict links between the representation 15 dimensions and coping procedures. Importantly, the adopted coping procedures will make 16 'common sense' to the individual in that they follow from their lay representations, regardless 17 of the type and effectiveness of the coping procedure adopted (Leventhal et al., 1992). 18 While the common-sense model does not specify particular coping procedures, a 19 number of theoretically-derived coping dimensions have been employed that broadly 20 correspond to problem-or threat-focused and emotion-focused coping categories (Carver et 21 al., 1989;Folkman & Lazarus, 1988 Leventhal et al., 2011), studies have demonstrated that automatic activation of an illness 24 representation is associated with activation of relevant coping procedures (Henderson, Orbell, distress, and negative associations with outcomes such as physical and social functioning, and 1 psychological well-being. Perceived control was positively associated with functioning, 2 vitality and psychological well-being, and negatively related to disease state 1  The typical pattern of relations between representations, coping, and outcomes 8 derived from previous research neglects to account for the moderating influence of contextual 9 factors. For example, transactional models of stress and coping indicate that problem-focused 10 coping strategies may not be adaptive if adopted to manage illnesses that are not controllable 11 (Lazarus & Folkman, 1984). Emotion-focused strategies may be more effective in these 12 contexts as they help individuals manage negative emotional reactions caused by the 13 perceived illness threat. This was recognized by Leventhal et al.'s (1992) 'goodness-of-fit ' 14 hypothesis, which indicates that the effectiveness of coping strategies on recovery and illness 15 progression is context dependent. However, current meta-analytic data do not account for the 16 effects of these contextual factors (e.g., Dempster et al., 2015;Hagger & Orbell, 2003). 17 From a theoretical perspective, meta-analyses of studies of the common-sense model 18 suggest that representations that signal illness threat do not guide individuals to adopt 19 problem-focused coping strategies as hypothesized by   be related to coping procedures that are emotion-focused, or, at least, not problem focused, 1 such as avoidance or denial. From a lay beliefs perspective, such coping responses make 2 common sense; the immediate management of emotional upheaval is an important first step 3 in active illness management because emotional distress is likely to interfere with 4 individuals' capacity to formulate and engage in problem-focused coping strategies 5 . This interpretation notwithstanding, threat perceptions are likely to 6 guide problem-focused coping when the individual has a clear, concrete "reality-bound 7 picture of a knowable threat and not a mysterious, remote, uncertain, and infinitely 8 threatening view of the illness process" (Leventhal et al., 1980, p. 20). There is empirical 9 support for this hypothesis, indicating that threat perceptions (e.g., high perceived 10 consequences) are related to treatment-seeking coping behaviors such as medication 11 adherence (Brewer et al., 2002), adherence to medical advice (Karademas,Karamvakalis,& 12 Zarogiannos, 2009), and self-care behaviors (MacInnes, 2013). Contextual factors such as 13 illness familiarity and coherence may determine whether threat perceptions lead to problem -14 or emotion-focused coping. Current meta-analytic data, therefore, capture overall tendencies 15 to prioritize emotion management in response to the perceived threat posed by the illness, but 16 may not reflect instances when the threat informs problem-focused coping efforts. 17 From an empirical perspective, it is important to note that no prior meta-analytic 18 synthesis of relations among the constructs of the common-sense model has controlled for the 19 effects of the other representation, coping, and outcome dimensions since they rely solely on 20 zero-order relations. Given the consistently strong intercorrelations among representation 21 dimensions, particularly among the threat dimensions (Hagger & Orbell, 2003), it is likely 22 that there will be a considerable degree of shared variance among these constructs that will 23 lead to substantively attenuated effects of illness representation dimensions on coping and 24 outcome constructs in multivariate analyses relative to the zero-order correlations (e.g., Evans Central to the common-sense model is the assertion that individuals' cognitive and 7 emotional representations of an illness threat will motivate a coping response to mitigate the 8 threat and related distress. Conceptually, therefore, specific coping procedures are proposed 9 to mediate the representation-outcome relations in the model ( Figure 2). The inclusion of 10 multiple representation dimensions along with numerous types of coping responses and 11 illness outcomes means that specific effects of each illness representation on illness outcomes 12 through coping strategies can be proposed and tested. From an analytic perspective, therefore, 13 illness representations "may exert effects [on illness outcomes] by eliciting or suppressing 14 both adaptive and maladaptive coping responses" (Gould et al., 2010, p. 635). 15 The accuracy of the process model can be evaluated on two levels. Taking a 16 generalized perspective, evaluating the extent to which coping mediates relations between 17 representation dimensions and illness outcomes will provide a test of the sufficiency of the 18 model. This would be a substantial advance on previous meta-analytic tests of the model 19 which have relied exclusively on zero-order correlations among constructs, a sub-optimal 20 approach to testing unique effects in social cognitive models (Hagger, Chan, et al., 2016). 21 Empirically, examination of the total indirect effects of all representation dimensions on 22 illness outcomes with all coping constructs as multiple mediators affords an assessment as to 23 the extent to which coping fully accounts for the impact of representations on outcomes. An 24 alternative hypothesis would be that a total indirect effect is present, but direct effects also exist, such that representation dimensions have unique effects on the outcome that are not 1 explained by coping. However, if the total indirect effects are found to be trivial or zero in the 2 presence of the direct effects, it would raise serious questions as to the adequacy of the 3 process model and could be grounds for its rejection. 4 The process model also permits a more fine-grained evaluation of the specific coping 5 mediated pathways by which individual representation constructs relate to outcomes. At this 6 level, testing for specific indirect effects may provide an indication of pathways not evident 7 when observing the total indirect effects, and may be 'missed' in an evaluation of the model. 8 For example, illness representations signaling threat (e.g., beliefs that an illness is highly 9 symptomatic, will have serious consequences, and has a chronic timeline) may compel an 10 individual to take action to attenuate the threat, consistent with Leventhal et al.'s (1980) 11 original hypothesis. Such actions may lead to adaptive improvements in illness outcomes . 12 However, the same representations may also lead an individual to engage in coping strategies 13 to manage negative feelings evoked by the perceived threat, such as emotion venting or 14 avoidance. Such strategies may not lead to improvements in disease state and functioning, 15 although they may make the person feel better. Illness identity, consequences, and timeline 16 may have two sets of specific indirect effects on illness outcomes through different coping 17 strategies that are not identified by observation of the total indirect effects alone. 18 The specific mediated pathways by which illness perceptions relate to outcomes 19 through coping procedures have seldom been explored in research on the common-sense 20 model. A few previous tests provide some preliminary evidence for the mediation hypothesis adherence, a problem-focused coping strategy. Such findings illustrate the potential for 8 mediation analyses to reveal process-related effects not evident when observing total indirect 9 effects or zero-order relations among illness representations, coping strategies, and illness 10 outcomes. 11 To date, tests of the process model are limited in that they were selective in the illness 12 representation and coping dimensions included in their mediation analysis and were restricted from studies measuring illness representations, coping strategies, and illness outcomes in 7 patient groups with chronic illnesses or conditions 2 . A critical mass of studies now exists 8 making a model test based on a cumulative synthesis of the available literature feasible. Our 9 analysis will extend previous research by employing a full matrix of correlations, providing a 10 test of model sufficiency that takes account of shared variance among model constructs, 11 identifying specific indirect effects, and evaluating contextual moderators of model relations. 12 Zero-order intercorrelations. The starting point of our analysis is to conduct a meta-13 analytic synthesis of studies on chronic conditions or illnesses. Across studies, we will 14 examine the zero-order, univariate patterns of relations among the three sets of variables that 15 constitute the process model: illness representation dimensions, coping strategies, and illness 16

outcomes. 17
Testing the sufficiency of the process model. We will subsequently subject the 18 corrected zero-order correlation matrix of the representation, coping, and outcome variables 19 to meta-analytic path analysis (Hagger, Chan, et al., 2016) to test the proposed mediation 20 effects in the process model. The hypothesized model is depicted in Appendix A 21 (supplemental materials). For the sake of parsimony, six separate path analyses of the process 22 2 Although there is no universal definition of a chronic illness or condition, there is general consensus on the characteristics that constitute chronic illness (Goodman, Posner, Huang, Perekh, & Koh, 2013). For the purposes of the current research, we define a chronic illness as a departure from a state of physical or mental well-being lasting more than three months (US Department of Health and Human Services, 2011) that likely requires ongoing medical care and limits activities of daily living (US Department of Health and Human Services, 2013). model are proposed, one with each illness outcome (disease state, distress, physical 1 functioning, psychological well-being, role functioning, social functioning) as the dependent 2 variable. In each analysis, the illness representation dimensions will be set as predictors of the 3 coping constructs, and each coping construct will be set as a predictor of the outcome. The 4 coping variables will serve as multiple mediators of each illness representation dimension on 5 the outcome. To test the sufficiency of the model, we propose a generalized hypothesis in 6 which the illness representation dimensions have statistically significant non-zero total 7 indirect effects on each illness outcome mediated by the coping constructs with no direct 8 effects of the representations on outcomes. An alternative model is also envisaged in which 9 representations are related to outcomes via two pathways: directly, and indirectly through 10 coping. In this case, the direct relations between representations and outcomes would be 11 attributable to mediators other than the identified coping constructs. 12 Testing model pathways. The analysis will also permit examination of specific 13 indirect coping pathways which might be missed by examining the total indirect effects 14 alone. In particular, we sought to address inconsistent evidence that threat representations (e.g., reduced distress, better functioning) and maladaptive (e.g., greater distress, increased 20 illness state) outcomes through specific coping strategies. These pathways may not have been 21 observed previously because researchers have not tested comprehensive models in which the 22 indirect effects of all representation dimensions on illness outcomes are specified through 23 multiple coping constructs. 24 Testing for moderators. Previous meta-analyses have observed considerable 1 coherence, greater perceptions of chronicity and emotional representations, increased 1 likelihood of attributing symptoms to the condition, and heightened concerns of the impact of 2 the illness on life. Such perceptions may lead to emotion-focused or avoidant coping 3 responses (e.g., catastrophizing thoughts, denial, hypervigilance), which likely lead to 4 outcomes that are maladaptive from a medical model perspective (Moss-Morris, Spence,& 5 Hou, 2011). We therefore expect illness representation dimensions signaling threat to be 6 related to emotion-focused coping strategies such as emotion venting among patients with 7 medically-unexplained symptoms. Similarly, we expect stronger links between these 8 dimensions and maladaptive outcomes such as increased distress and lower functioning in 9 patients with medically-unexplained symptoms. 10 Stage of illness is a further viable moderator. Consistent with model hypotheses, the 11 impact of representations on coping strategy selection is expected to vary in accordance with 12 whether the illness is in an early-stage diagnosis or whether the diagnosis is distal and 13 experience with the condition is more extensive. As threat perceptions and adverse emotional 14 responses are likely to be elevated on initial diagnosis, we expect stronger effects of threat 15 perceptions (consequences, timeline, and identity) and emotional representations on emotion-16 focused coping strategies, and weaker effects on problem-focused coping strategies, in 17 studies of recently-diagnosed patients relative to those on patients for whom some time had 18 elapsed since diagnosis. As patients receive more information and have greater experience 19 with symptoms and treatment over time, they may adopt more problem-focused coping 20 strategies in response to threat-related and control representations. Testing for the moderation 21 effects of illness stage is therefore likely to capture a dynamic process in the model in which 22 coping responses change with illness and treatment experience (Leventhal et al., 2016

Included Studies and Characteristics 13
The literature research identified 333 articles that met inclusion criteria on initial 14 screening (Appendix C, supplemental materials). A substantial proportion of the eligible 15 articles (k = 172) did not report the necessary data for the analysis. The corresponding authors 16 of these articles were contacted to obtain additional data. We were unable to source the 17 unreported data for 39 articles, and several articles either reported data from multiple samples 18 within a single study or comprised overlapping samples (multiple studies using the same 19 data). Details of articles providing multiple and overlapping samples are provided in Tables 1  20   and 2  included studies including sample sizes, study design, demographic details, illness types 23 encompassed by the studies, moderator coding, and additional variables are provided in Table  24 conditions with forms of cardiovascular disease (k = 36), diabetes (k = 32), cancers (k = 22), 1 arthritis (k = 16), forms of chronic pain (k = 12), chronic obstructive pulmonary disease (k = 2 9), end-stage renal disease (k = 9), chronic fatigue (k = 7), multiple sclerosis (k = 7), irritable 3 bowel syndrome (k = 6), psoriasis (k = 6), and hypertension (k = 4) the most frequently cited. 4 Studies were largely focused on older samples (median of the average sample age reported in 5 studies = 52 years) with most studies having approximately equal ratios of females and 6 males. Single sex samples were usually due to studies targeting illnesses that generally affect, 7 or occur exclusively in, females (e.g., cervical cancer) or males (e.g., prostate cancer  on the basis of content and label into the six illness representation  2   dimensions: control, consequences, identity, timeline, emotional representations, and illness  3 coherence. For instance, the seriousness, control, and symptoms dimensions from Hampson 4 et al.'s (1990) personal models inventory were coded as equivalent to the consequences, 5 control, and identity dimensions from the common-sense model. Similarly, the DIRQ (2003) 6 perceived threat to health and perceived impact dimensions were coded as equivalent to the 7 consequences dimension, and the effectiveness to prevent future complications and 8 effectiveness to control diabetes dimensions were coded as equivalent to the control 9 dimension. In order to provide external validity for our coding, the coding scheme was 10 confirmed by three raters who independently conducted the classification procedure based on 11 formal definitions of the six illness representation dimensions. Inter-rater reliability indicated 12 perfect agreement among the raters (Cohen's κ = 1.00). 13 Many of the included studies used modified versions of the IPQ-R to assess the 14 specific illness or condition of interest. In addition, the IPQ-R was translated, validated and 15 standardized for various populations. Studies followed standard IPQ-R scoring procedures 16 such that higher scores on the consequences, identity, timeline, and emotional representation 17 dimensions reflected more serious consequences, greater symptom frequency, chronic 18 duration, and heightened negative emotional reactions. In contrast, high scores on the control 19 and coherence dimensions represented greater perceived control over, and a clearer 20 understanding of, the illness. 21

Classification of Coping Strategies 22
The present sample of studies used a large number of instruments to measure coping 23 strategies, many of which were based on cognitive-motivational-relational models of stress 24 and coping (e.g., Carver et al., 1989;Lazarus & Folkman, 1984). A key task in the present meta-analytic synthesis was to classify measures used to tap coping strategies in the current 1 sample and classify them into conceptually-distinct categories based on theory. Many of the 2 studies in the sample adopted generic, previously-validated questionnaires to assess coping 3 strategies, such as the COPE inventory (Carver et al., 1989) Following Hagger and Orbell's (2003) classification procedure, we identified six 8 distinct coping categories, five of which reflected the general scope and content of the pool of 9 generic coping scales, namely avoidance, cognitive reappraisal, emotion venting, problem-10 focused coping (generic strategies), and seeking social support (see Appendix F, 11 supplemental materials, for classification of constructs from generic coping measures into 12 coping categories). A sixth coping category, labelled problem-focused specific coping 13 strategies, encompassed active attempts to address the illness by means of specific illness-14 related coping behaviors, such as medication and dietary adherence, illness-related self-care 15 behaviors, and attendance at illness-related medical appointments (see Appendix G, 16 supplemental materials, for the classification of non-generic coping measures into coping 17 strategies). We employed three independent raters with expertise in health psychology and 18 theories of stress and coping to code measures of coping into the six a priori categories based 19 on formal definitions of the six coping categories and the content of the source items of the 20 scales used to tap the constructs (see Appendixes F and G, supplemental materials). Inter-21 rater reliability coefficients indicated good agreement between the raters for each coping 22 dimension category (Fleiss-corrected κ = .91). Differences in classifications were discussed 23 among the raters with respect to the category definition and item content with resolution 24 based on consensus between all three raters.

Classification of Illness Outcomes 1
Given the range of measures used to tap illness-related outcomes in the current 2 sample of studies, it was also important to categorize illness outcome measures into distinct 3 categories. As with the coping strategies, we adopted Hagger and Orbell's (2003) procedure 4 to code the outcome measures into relevant categories. We identified six distinct illness 5 outcome categories a priori: disease state, physical functioning, psychological distress, 6 psychological well-being, role functioning, and social functioning. Although Hagger and 7 Orbell included a seventh category, vitality, this category was merged with the psychological 8 well-being category in the present analysis due to considerable overlap in the content of 9 measures used to tap constructs across these categories. While many of the studies adopted

Meta-Analytic Strategy
The effect size metric of interest in the present meta-analysis was the zero-order 1 correlation coefficient. We conducted separate meta-analyses of intercorrelations among the 2 illness representation and coping strategy dimensions, and of correlations between the 3 representation, coping and outcome dimensions, resulting in 138 effect sizes 4 . Hunter and 4 Schmidt's (2004) formulas were adopted to correct the effect sizes for statistical artefacts. 5 The Hunter and Schmidt approach is equivalent to a random effects model for meta-analysis 6 and is considered optimal as it provides estimates that are generalizable to the population 7 rather than to the sample of included studies alone (Field, 2001;Hagger, 2006;Hunter & 8 Schmidt, 2000). We corrected the effect sizes for both sampling and measurement error and 9 utilized the zero-order correlation coefficient as the effect size metric. The meta-analyses 10 were conducted using the MetaQuick (Stauffer, 1996) and Comprehensive Meta-Analysis 11 Version 2 (Borenstein, 2011) statistical software. In the event that a particular category of our 12 key variables was represented by more than one construct in a specific study (e.g., the 13 positive reintegration and acceptance scales from the COPE inventory were both classified 14 into the cognitive reappraisal coping category), the average of the correlation coefficients was 15 taken to provide a single test of the expressed relationship for use in the meta-analysis 16 consistent with Hunter and Schmidt's (2004) methods. Studies reporting effect size data in 17 other metrics (e.g., standardized mean difference) were converted to correlation coefficients. 18 We corrected for measurement error using Cronbach alpha reliability coefficients of the 19 constructs used in each effect size calculation. Where reliability statistics were unavailable, 20 measurement error was inferred from available attenuation statistics using Stauffer's (1996) 21

formula. 22
The analyses provided key summary statistics of the effects among illness 23 representations, coping strategies and illness outcomes: the uncorrected averaged correlation 24 coefficient for the effect (r), the averaged correlation corrected for sampling error only (r+), 1 the fully-corrected averaged correlation coefficient corrected for both sampling and 2 measurement error (r++), 95% confidence intervals (CI95) of the fully-corrected averaged 3 correlation which provided a formal test of statistical significance for each effect, and the 4 variance in the fully-corrected averaged correlation accounted for by the statistical artefacts 5 of sampling and measurement error. If the vast majority of the variance in the averaged effect 6 size can be accounted for by statistical artifacts -Hunter and Schmidt (2004) recommend a 7 75% cut-off criterionthen the effect should be considered homogenous (i.e., free from bias 8 other than sampling and measurement error). If the proportion of the variance falls below this 9 criterion, then it is likely that there is substantial variance in the effect size across the studies 10 that cannot be attributed to methodological artifacts and indicates the possibility that 11 additional variance in the effect exists across the studies that can be attributed to extraneous 12 or 'moderator' variables. We also report Cochran's (1952) Q statistic, which provides a 13 formal test of the hypothesis that variation in effect sizes across studies is greater than that 14 expected on the basis of sampling error alone. Given that the number of studies (k) varies 15 across meta-analyses, the Q-statistic cannot be compared across analyses. An alternative is 16 offered by the I 2 statistic and its confidence interval (Higgins & Thompson, 2002). The a search for extraneous moderators of the effect. Reporting the between-study or heterogeneity variance (τ 2 ) for each averaged effect size is also recommended. Small values 1 for τ 2 are indicative of low heterogeneity in the effect size across studies. Finally, we 2 evaluated the presence of small-study bias in the sample of effect sizes by computing 3 statistics based on plots of the effect size from each study against study precision (usually the 4 reciprocal of the study sample size). Asymmetry in the predicted 'funnel' shape of the plots is 5 considered evidence of small study bias, that is, the tendency for studies included in the 6 analysis to exhibit large effects relative to their size. This is often interpreted as a potential 7 indicator of publication bias and the tendency for journals to favor statistically significant 8 findings in underpowered studies. Specifically, we used Egger, Davey Smith, Schneider, and 9 Minder's (1997) regression test, which indicates the extent to which effect sizes are predicted 10 by study precision and Duval and Tweedie's (2000) 'trim and fill' technique, which identifies 11 studies that deviate from the expected shape of the 'funnel' plot and adjusts the plot for 12 missing studies to make it more symmetric. 13

Moderator Coding and Analyses 14
Assuming substantive, non-trivial variability in the effect sizes of the relations in the 15 illness representation, coping, and outcome constructs unattributed to the statistical artifacts 16 we corrected for in the model, we aimed to examine the effect of moderators of the model 17 effects. A series of candidate moderator variables was identified and coded in the current 18 analysis: study design, illness type, medically-explained versus medically-unexplained 19 symptoms, illness stage, and study methodological quality. Moderator coding is presented in 20 Table 3 (Appendix E, supplemental materials). We evaluated the effects of the moderator on 21 correlations between the illness representation, coping, and outcome variables across studies 22 by conducting separate meta-analyses for each level of the moderator 5 . Differences in the 23 5 Raw data files used to conduct the moderator analyses for each effect including moderator coding are available online at https://osf.io/g48nt/ corrected correlations across moderator groups were evaluated by inspection of the 95% 1 confidence intervals about each correlation with Welch's t-test providing a formal test of 2 difference. 3 Study design. Study design was coded as cross-sectional, longitudinal, or intervention. 4 Studies measuring all study variables at a single point in time were coded as cross-sectional. 5 This included studies adopting longitudinal or intervention designs that only reported 6 correlational data at a single time point or independent of the intervention. Studies adopting 7 designs in which either coping or outcomes were measured at a distal point in time relative to 8 the illness representation dimensions were coded as longitudinal. Studies including an 9 intervention or experimental component targeting change in illness representations, and 10 change in a subsequent behavior or illness-related outcome, were coded as intervention. 11 However, these studies numbered very few (3.2%) and, in many cases, data were unavailable 12 (1.5%), which precluded moderator comparisons between groups of studies adopting 13 intervention or experimental designs and groups of studies with cross-sectional and 14 longitudinal designs. 15 Illness type. Coding of illness type was based on the description of the illnesses 16 reported in the study. Studies were classified as having cancer as the target illness if study 17 participants had received a cancer diagnosis or were undergoing cancer screening. Studies 18 were classified as having diabetes as the target illness if the patient group comprised either 19 type I or type II diabetics. Studies were classified as targeting cardiovascular disease if the 20 patient group was described as having experience with, or diagnosis for, a form of 21 cardiovascular disease including acute coronary syndrome, angina, atrial fibrillation, cardiac 22 chest pain, coronary artery/heart disease, heart failure, hypercholesterolemia, myocardial 23 infarction, and stroke. In the case of each illness type moderator, the contrasting moderator 24 group comprised the remaining studies in which the target illness was absent.

Medically-explained versus medically-unexplained symptoms. Illnesses and 1
conditions were classified as those with medically-explained symptoms and medically-2 unexplained symptoms by two independent researchers with research experience in 3 behavioral medicine. Studies describing the target illness or condition of the study as having 4 unknown or uncertain etiology, or if the symptoms of the illness or condition are known to 5 have unknown or uncertain causes, were classified as having medically-unexplained 6 symptoms with the remaining illnesses classified as having medically-explained symptoms. 7 Inter-rater reliability for the coding of studies for medically-explained and medically-8 unexplained illnesses resulted in good inter-rater reliability (intraclass correlation = .83; CI95 9 = .78, .86). Inconsistencies were resolved through discussion and consultation with medical 10 definitions of the reported conditions alongside the description of the illness and patient 11 group reported in the studies. 12 Illness stage. Illness stage was coded as time since diagnosis. Studies on patients who 13 received their diagnosis at four weeks or earlier prior to study data collection were classified 14 as recently diagnosed while studies on patient groups who had received their diagnosis more 15 than four weeks prior to data collection were classified as having a non-recent diagnosis. A 16 third category was identified for samples comprising both patients who had received a recent 17 diagnosis and patients who had not received their diagnosis recently. 18 Study methodological quality. Study quality was assessed using a methodological 19 quality checklist developed for the current study. Given that the current sample comprised 20 exclusively of studies adopting survey methods and largely adopted correlational designs, our 21 checklist was based on those developed for these types of study.  Table 4 (Appendix I, supplemental materials). Studies meeting 6 quality standards were assigned a score of 1 for each criterion and those not meeting the 7 quality standard or provided insufficient information to evaluate the criterion were assigned a 8 score of zero. Cross-sectional studies could achieve a maximum score of 13 and longitudinal 9 studies a maximum score of 16. We weighted the scores for purposes of comparison by 10 dividing raw scores by the number of relevant items and multiplying by 10. 11 Studies were scored on the checklist by a researcher with experience in the use of 12 methodological quality checklists 6 . One study could not be scored because the relevant data 13 were not available and was omitted from subsequent analyses. A subset of the sample of 14 studies (N = 20) was independently scored by two other researchers with high agreement 15 between raters across items (mean agreement = 94.17%) and inter-rater reliability (mean 16 Fleiss' κ = .82). Inconsistencies were resolved through discussion and attributed to minor 17 interpretation errors of the quality standard for two criteria. The criteria were subsequently 18 revised on the basis of the discussion and applied to the coding of the entire sample. We 19 coded two methodological quality categorical variables for use in our analysis. After a tertile 20 division of the studies by checklist score, studies with scores in the upper third were 21 classified as high quality and those in the lower third classified as low quality 7 . This strategy 22 allows comparisons of studies with extreme scores but also reduces the sample size. We 1 therefore coded an additional moderator variable based on a cut-off checklist score ( analysis used as the input matrices 8 . In order to minimize complexity, we estimated separate 10 path analyses for each of the six illness outcomes. Each model included the six illness 11 representation dimensions as exogenous independent predictors of the illness outcome 12 mediated by multiple pathways through the coping strategies. We tested two models for each 13 outcome. The first (Model 1) assumed that the effects of the illness representations were fully 14 mediated by the coping strategies, and, as such, no direct pathways were freed between the 15 illness representation dimensions and illness outcome. This model assumes that the coping 16 In addition, we conducted sensitivity analyses to illustrate the extent to which the 19 large sample sizes in the current meta-analysis affected the precision and statistical 20 significance of the parameter estimates in our models, and whether these changes affected our 21 inferences. In the sensitivity analyses, our models were re-estimated with substantively 22 smaller sample sizes. Specifically, we used the largest sample size of the included studies (n 23 = 3,130) and a sample size that approximated the average of the five next-largest sample sizes 24 of the included studies (n = 1,000) as input sample sizes. Given that variability estimates and significance tests associated with the effect sizes are influenced by sample size, we expected 1 the width of the 95% confidence intervals of the parameter estimates in the models to 2 progressively increase, and that smaller effects would be more likely to become statistically 3 non-significant, with declining sample size. In such cases, the estimated effect may be 4 rejected even if it is not trivial in size. 5 We also conducted additional sensitivity analyses comparing model effects across 6 levels of the candidate moderator variables. We estimated the proposed models (Model 1 and 7 Model 2) using the meta-analyzed mean correlation matrices in each moderator group. We 8 compared overall fit of model across levels of the moderator using the AIC to determine 9 whether the moderator affected our determination of which model exhibited best fit derived 10 from the full-sample analysis. We also report the CFI for each model, with absolute values 11 for Model 1 illustrating the level of misspecification in the model when direct effects were 12 fixed to zero. In addition, we tested whether the pattern of effects in each model varied across 13 moderator levels using multi-group path analysis. In these analyses, paths among the illness 14 representation, coping, and outcome variables were set to be invariant across the models 15 estimated at each level of the moderator by imposing a set of equality constraints. We 16 conducted the multi-group analysis for Model 1 and Model 2, with additional constraints 17 specified in Model 2 to test for invariance in the direct effects of the illness representation 18 dimensions on the outcome variable. Adequacy of the multi-group models was evaluated 19 using multiple recommended criteria for goodness-of-fit: the CFI, the normed fit index (NFI), 20 and the root mean square error of approximation (RMSEA), with values exceeding .95 for the 21 CFI and NFI and .050 or less for the RMSEA. Incorrectly-imposed constraints indicating 22 paths that are non-invariant across levels of the moderator will result in misspecification in 23 the model fit. As with the path models estimated on the full sample of studies, sensitivity analyses were conducted separately for each of the six outcome variables (disease state, 1 psychological distress, psychological well-being, and role, social, and physical functioning). 2

Corrected correlations 4
Averaged correlation coefficients corrected for sampling and measurement error for 5 all variables in the meta-analysis are appear in Table 5 in Appendix J (supplemental 6 materials) with confidence intervals, variability and heterogeneity statistics, and analyses 7 based on funnel plots. Intercorrelations among illness representation dimensions yielded a 8 pattern consistent with results of previous meta-analytic findings (Hagger & Orbell, 2003). 9 Specifically, the identity, serious consequences, timeline, and emotional representation 10 dimensions were statistically significantly and positively related to each other, and 11 significantly and negatively related to the perceived control and illness coherence 12 dimensions. The only exception to this pattern was the coherence-timeline relationship, 13 which was not statistically significant. Perceived control and illness coherence were 14 significantly and positively correlated. 15 Relations between illness representations and coping strategies, and between coping 16 strategies and illness outcomes, were also consistent with previous analyses. The identity, 17 serious consequences, and timeline dimensions were statistically significantly and positively 18 associated with emotion venting and avoidance coping strategies, and not significantly related 19 to the cognitive reappraisal, problem-focused generic, and problem-focused specific coping 20 strategies. Emotional representations were significantly and positively related to the emotion 21 venting and avoidance coping strategies, and negatively related to cognitive reappraisal and 22 problem-focused generic strategies. Analogously, the identity, consequences, timeline, and 23 emotional representation dimensions were significantly and positively correlated with disease 24 state and psychological distress, and significantly and negatively related to physical, role, and social functioning, and psychological well-being. Perceived control and illness coherence 1 were significantly and positively related to cognitive reappraisal, problem-focused generic, 2 problem-focused specific, and seeking social support coping strategies, and significantly and 3 negatively related to avoidance, with the exception of the relation between coherence and 4 social support. Similarly, control and coherence were significantly and positively associated 5 with physical, role, and social functioning, and psychological well-being, with the exception 6 of the correlation between coherence and social functioning. 7 Focusing on the heterogeneity statistics, according to both Hunter and Schmidt's 8 (2004) 75% rule and Cochran's Q statistic, all but 13 of the 138 effects were found to have a 9 statistically significant degree of heterogeneity across studies, indicating the likely presence 10 of moderator variables acting on the observed relationships. This was corroborated by the I 2 11 statistic, which showed at least moderate heterogeneity for the majority of the effect sizes 12 appear to point to deviations from the expected pattern in funnel plots and potential small 17 study bias for some of the effects, the high degree of heterogeneity is a cause for concern 18 given that problems have been identified with interpreting these statistics in the presence of Findings indicating substantial bias using these tests in the current analysis should, therefore, 21 be interpreted in light of this concern. 22

Moderator Analyses 23
We examined the effect of moderators on correlations among the illness 24 representation, coping strategies, and illness outcome constructs across studies by conducting our meta-analysis at each level of the candidate moderators: study design, illness type, 1 medically-explained and medically-unexplained symptoms, illness stage, and methodological 2 quality. Averaged corrected correlations, confidence intervals, and heterogeneity statistics 3 from the moderator analyses are presented in Table 6 in Appendix K (supplemental 4 materials). In all cases, we found little evidence for moderator effects. Of the 138 effects 9 5 tested in each analysis, fewer than seven effects per moderator were significantly different 6 across levels of the moderator and there was substantive overlap in confidence intervals in the 7 vast majority of the tests. In addition, the analyses did little to resolve the heterogeneity in the 8 effect sizes with moderate-to-high heterogeneity according to I 2 values and significant values 9 for Q observed for effects within each moderator group. Of the correlations that did exhibit 10 statistically significant differences, many included moderator groups with low numbers of 11 studies (k < 10); in such cases the confidence intervals, t-tests, and heterogeneity tests may 12 not be reliable. Overall, there was little indication of systematic variation in effect sizes 13 attributable to the candidate moderators in the current analyses. 14

Path Analyses of the Process Model 15
Sufficiency of the process model derived from the common-sense model (Appendix 16 A, supplemental materials) was tested using a series of mediated path-analytic models using 17 the averaged corrected correlation matrices among the illness representation, illness outcome, 18 and coping constructs as input 10 . Separate analyses for each illness outcome construct 1 (disease state, psychological distress, physical functioning, psychological well-being, role 2 functioning, and social functioning) were conducted to minimize model complexity. 3 Model sufficiency. We estimated two models to test the sufficiency hypothesis. 4 Model 1 specified direct effects of the illness perception dimensions (identity, consequences, 5 timeline, control, illness coherence, and emotional representations) on coping constructs 6 (avoidance, cognitive reappraisal, emotion venting, specific and generic forms of problem-7 focused coping, and seeking social support), direct effects of the coping constructs on the 8 illness outcome, and indirect effects of all the illness perceptions constructs on the illness 9 outcome mediated by the coping constructs. Model 2 was identical to Model 1 but also 10 included direct effects of the illness representations dimensions on illness outcomes. 11 Goodness-of-fit statistics for each of the six models are provided in Table 7  for relations between the illness representation and outcome variables. We therefore rejected 18 10 Correcting for methodological artifacts is known to increase the magnitude of effect sizes, perhaps more than is appropriate, which may lead to over-or under-estimation of the true effect sizes (see Johnson & Eagly, 2014; Köhler, Cortina, Kurtessis, & Gölz, 2015). For example, interdependency between the reliability estimates of predictor and criterion variables of a particular effect could effectively lead to 'double correction' for measurement error (Köhler et al., 2015). Such corrections may lead to an inflation of effect sizes and increase the probability of finding statistically significant effect sizes. In order to ascertain the extent to which interdependency of reliabilities was a problem, we followed the recommendations of Köhler et al. and computed correlations between the predictor and criterion reliability estimates used to correct our effect sizes for measurement error. Results revealed that the correlations were generally small (median = .25, inter-quartile range = .37), which indicates that there may be some bias attributable to correction for this artifact but it is unlikely to be substantive in most cases. We also reestimated our models using the uncorrected averaged correlations and compared the results to the analyses using the corrected correlations (see Table 5, Appendix J, supplemental materials). Results are presented in Tables 21 to 23 in Appendix O (supplemental materials). Unsurprisingly, the analysis demonstrated that the effects for the analyses using raw data tended to be smaller than those for the corrected data. However, the pattern of effects remained in both sets of analyses, indicating that the corrections tended to alter the strength rather than pattern of effects.

Model 1 in favor of Model 2. Subsequent examination of direct and indirect effects in Model 1
2 was warranted to evaluate the extent to which the coping variables mediated the illness 2 representation-illness outcome relations. Parameter estimates for the direct (Table 8), indirect 3 (Table 9), and total (combined direct and indirect effects; Table 10) effects for each model 4 are provided in Appendix L (supplemental materials). 5 Direct effects. Direct effects of illness representation dimensions on coping strategies, 6 of coping strategies on illness outcomes, and of illness representations on outcomes provided 7 indication of the unique predictors of the coping and illness outcome variables while 8 simultaneously accounting for the effects of other constructs in the model (see Table 8). 9 Focusing first on the direct effects of the illness representation dimensions on illness 10 outcomes, we found statistically significant, non-trivial effects of perceived consequences 11 and illness identity on all outcomes, with the exception of the consequences-distress effect 12 which was small by comparison. Effects were positive for disease state and distress and 13 negative for the functioning outcomes and well-being. Perceived control had positive non-14 trivial direct effects that were statistically significant and greater than .10 on physical, social, 15 and role functioning, and well-being, while effects on disease state and distress were smaller 16 and trivial in effects size. The pattern of direct effects for these representation dimensions on 17 outcomes followed the same pattern as the zero-order correlations among these constructs. A 18 notable exception was the direct effect of control on disease state which was positive in the 19 models but had a negative zero-order correlation. Emotional representations had statistically 20 significant non-trivial negative direct effects on well-being and role functioning, and positive 21 effects on distress and social functioning. The timeline dimension had statistically significant 22 non-trivial positive direct effects on role functioning and well-being. Effects for emotional 23 representations and timeline deviated from the pattern of effects in the zero-order 24 correlations: these dimensions were negatively correlated with functioning.
Focusing on the direct effects of the illness representation dimensions on coping, we 1 found statistically significant non-trivial effects of the representation dimensions on the 2 coping constructs. Dimensions with the largest effects were emotional representations which 3 positively predicted avoidance, emotion venting, and seeking social support strategies, and 4 negatively predicted problem-focused generic coping and cognitive reappraisal. Control had 5 statistically significant non-trivial positive direct effects on generic and specific forms of 6 problem-focused coping, cognitive reappraisal, and social support. There were also 7 statistically significant non-trivial positive direct effects of identity on avoidance, cognitive 8 reappraisal, and emotion venting. Other effects of the illness representation dimensions on 9 coping constructs were smaller and trivial by comparison. 10 Direct effects of coping strategies on illness outcomes tended to be larger than the 11 effects of the illness representation dimensions on coping, and were consistent with the 12 theory-derived predictions from the process model and previous analyses (Hagger & Orbell, 13 2003). We found statistically significant non-trivial direct effects of avoidance on all 14 outcomes, with positive effects on disease state and distress, and negative effects on physical, 15 role, and social functioning, and well-being. A similar pattern of direct effects was found for 16 emotion venting, the only exception was that emotion venting negatively predicted disease 17 state when the correlation between these variables was not significant, which may be 18 indicative of a suppressor effect. Problem-focused generic coping had statistically significant 19 non-trivial negative direct effects on disease state, distress, and role functioning, and positive 20 effects on physical functioning and well-being. Problem-focused specific coping had 21 statistically significant non-trivial positive direct effect on role functioning, with smaller 22 trivial effects on other outcomes. 23 Overall, findings of the direct effects indicate that illness representations had 24 statistically significant non-trivial direct effects on illness outcomes, consistent with the better overall fit of the model that included direct effects (Model 2) relative to the model that 1 assumed no direct effects (Model 1). Effects for the coping strategies on illness outcomes 2 tended to be larger compared to the effects of the illness representation dimensions on coping 3 constructs. 4 Total indirect effects. While we rejected our hypothesis that the effects of illness 5 representations on illness outcomes would be completely mediated by coping strategies, this 6 conclusion did not rule out the possibility of indirect effects consistent with hypotheses of the 7 common-sense model. Examination of the total indirect effects to establish the extent to 8 which the effects of illness representation dimensions on illness outcomes were mediated by 9 the coping strategies was warranted . Total indirect effects are 10 presented in Table 9 (Appendix L, supplemental materials). 11 We found statistically significant non-trivial positive total indirect effects of perceived 12 control on physical functioning, role functioning, and psychological well-being, and negative 13 effects on psychological distress and disease state. There were statistically significant non-14 trivial negative total indirect effects of identity on physical, role, and social functioning, and a 15 positive effect on distress. Chronic timeline had statistically significant non-trivial positive 16 indirect effects on psychological social functioning, physical functioning, and well-being, and 17 negative effects on distress and disease state. We found statistically significant non-trivial 18 positive total indirect effects of emotional representations on disease state and distress, and 19 negative total indirect effects on physical functioning, social functioning, and well-being. 20 Other indirect effects were small and trivial by comparison. Overall, these effects provided 21 support for the indirect, coping-mediated effects of illness representation dimensions on 22 illness outcomes consistent with the process model. 23 We also estimated the mediation proportion for each total indirect effect (PM), which 1 reflects the proportion of the total effect accounted for by the indirect effect 11 . This is directly 2 relevant to the evaluation of model sufficiency as it provides an indication of the extent to 3 which the total indirect effects of illness representations on outcomes through coping 4 contribute to the total effect. A substantial PM value indicates that the mediated pathway 5 makes a viable contribution to explaining the link between representations and outcomes, 6 while a trivial value indicates that the mediated path was of little relevance relative to the 7 direct effect. Results revealed that many of the indirect effects accounted for substantial 8 proportions of the total effects of illness representations on illness outcomes 12 . Prominent 9 among these were the control, emotional representations, timeline, and coherence 10 dimensions, for which the total indirect effects accounted for substantive proportions of the 11 total effect. For the identity and serious consequences dimensions, the majority of the indirect 12 effects accounted for a trivial proportion of the total effect, with the notable exception of the 13 indirect effect of consequences on distress. Findings provide support for the process model, 14 verifying that the indirect effects make a substantive contribution to the overall effects in the 15

model. 16
Specific indirect effects. Our analyses also enabled us to isolate specific mediated 17 effects involving each of the illness representation, coping, and outcome variables that 18 constituted the total indirect effects. Specific indirect effects are presented in Table 9 19 11 One of the limitations of the PM statistic is that it is difficult to interpret when total effect comprises negative and positive direct and indirect effects, such as the effect of perceived control on disease state in the current analysis which was made up of a positive direct effect (β = .034) and a negative total indirect effect (β = -.079). The total effect (β = -.045), therefore, represents the sum of the direct and total indirect effects. In this case, the positive direct effect has the effect of attenuating the negative indirect effect. However, as the total effect serves as the denominator for the PM calculation, it may give misleading scores for the PM when the total effects comprise negative and positive effects, as in the previous example. This may lead, for example, to the PM exceeding unity such that it does not represent a true proportion (Preacher & Kelley, 2011). A solution was to estimate PM using the modulus of the total effect in its calculation. The PM, therefore, reflects the proportion of the total effect accounted for by the indirect effect regardless of whether the combination of the direct and indirect effects leads to an attenuation of the total effect due to the combination of negative and positive scores. 12 Although no published guidelines exist, we considered PM values > .25 to be of substantive value as it corresponds to a proportion of the total effect accounted for by the indirect effect above the 25 th percentile.
(Appendix L, supplemental materials). An interesting trend in the current findings was that 1 some of the total indirect effects of representation dimensions on outcomes were either zero 2 or relatively trivial in size. However, in some cases the total indirect effects comprised both 3 positive and negative specific indirect effects. These specific indirect effects were 4 approximately equal in magnitude but opposite in sign (i.e., positive and negative) leading to 5 a null or very and trivial small total indirect effect. A consideration of the pattern of specific 6 indirect effects, therefore, may reveal important information on the pattern of effects of the 7 illness perception variables on outcomes that cannot be gained from observing the total 8 indirect effects alone. The specific indirect effects may facilitate interpretation as to whether 9 the indirect effects of illness representation dimensions have effects on illness outcomes 10 considered adaptive (e.g., reduced distress, better functioning) or maladaptive (e.g., increased 11 distress and disease state), through coping procedures, or whether both adaptive and 12 maladaptive patterns of indirect effects through coping are present. We evaluate the pattern of 13 the specific indirect effects for each illness perception dimension on each illness outcome in 14 turn. 15 Perceived consequences had zero total indirect effects on disease state and physical 16 and role functioning, and the effect on well-being although statistically significant was very 17 small (β < .01). This was unexpected given that primary studies and meta-analyses have 18 consistently reported statistically significant, substantive negative correlations between 19 perceived consequences and functioning and well-being, and positive correlations with 20 distress and disease state. However, theory suggests that perceived consequences may also 21 have positive effects on adaptive illness outcomes by motivating individuals to take action to 22 mitigate the threat and there is previous evidence to support this (e.g., Brewer et al., 2002). 23 Based on this evidence, it is possible that both positive and negative specific indirect effects 24 would be present, and that these effects would amount to null or relatively small total indirect effects. Close inspection of the specific indirect effects revealed consistency in the pattern of 1 effects of consequences on each outcome via the coping constructs as mediators. Specifically, 2 consequences tended to predict maladaptive outcomes through avoidance coping and 3 adaptive outcomes through problem-focused coping. For example, consequences had 4 statistically significant and positive effects on disease state and distress through avoidance, 5 and significant negative effects on these outcomes through problem-focused generic coping. 6 The effects for consequences were corroborated by the mediation proportion statistics. 7 The specific indirect effects for consequences through avoidance and problem-focused 8 coping accounted for a substantial proportion of the overall indirect effect. There were also 9 significant negative effects of consequences on functioning and well-being through avoidance 10 and significant positive effects on these outcomes through problem-focused generic coping. 11 Together these effects were similar in magnitude and opposite in sign leading to the null or 12 relatively small total indirect effects of consequences on outcomes. The effects for the 13 consequences dimension, therefore, comprised both positive and negative indirect 14 associations with adaptive and maladaptive outcomes through coping. This is consistent with 15 the view of consequences as a representation of threat, which may motivate patients to take 16 action to deal with the threat or to engage in procedures to manage the concomitant emotional 17 upheaval. The specific indirect effects, therefore, reveal the multiple pathways by which 18 consequences impact on outcomes not indicated by the total indirect effect. 19 Similar positive and negative patterns of specific indirect effects were found for the 20 identity dimension on illness outcomes. Total indirect effects revealed statistically significant 21 effects for identity on distress, significant negative effects on physical, role, and social 22 functioning, and null effects on disease state and well-being. Decomposition of the specific 23 indirect effects revealed that the total indirect effect comprised effects of opposing sign. For 24 example, identity had significant positive indirect effects on disease state and distress through avoidance, and significant negative indirect effects on physical, role and social functioning, 1 and well-being through this mediator. Mediation proportion statistics indicated that a 2 substantial proportion of the indirect effect of identity on each outcome was through 3 avoidance. There were also statistically significant positive effects of identity on physical and 4 social functioning, and well-being mediated by problem-solving generic coping, although the 5 size of these effects was small and trivial by comparison. Analogously there were statistically 6 significant but small negative effects of identity on distress and disease state through 7 problem-focused generic coping. These oppositely-valanced effects had the effect of reducing 8 the size of the total indirect effects for this representation dimension. The total indirect effects 9 of identity on outcomes were, therefore, substantially weaker due to indirect effects of 10 opposing sign. Again, this is consistent with the view of identity as representing illness threat 11 and its potential to affect both adaptive and maladaptive outcomes through multiple 12 pathways. problem-focused coping. We also found significant negative specific indirect effects of control on disease state and psychological distress mediated by problem-focused coping and 1 cognitive reappraisal, effects that were in keeping with the predicted pattern. Overall, 2 perceived control was related to adaptive outcomes including improvements in functioning 3 and well-being and reductions in disease state and distress. 4 Illness coherence had statistically significant positive total indirect effects on role 5 functioning and well-being and negative indirect effects on distress and disease state. Such 6 effects indicate that individuals with a clearer understanding of the illness may be better 7 equipped to identify relevant coping strategies as they are likely to have relevant information 8 regarding which strategies may be more effective. Problem-focused coping and cognitive 9 reappraisal were, therefore, expected to be key mediators of effects of coherence on 10 outcomes. However, observing the specific indirect effects revealed that the effect of 11 coherence on outcomes was not mediated by problem-focused coping, with the exception of 12 the significant positive effect of coherence on role functioning through problem-focused 13 specific coping. Instead, the specific indirect effects of coherence on outcomes were mediated 14 by avoidance and emotion venting in most cases, and substantive proportions of the indirect 15 effect were accounted for by the specific effects through these mediators. For example, 16 coherence was statistically significantly and positively related to physical, role, and social 17 functioning through emotion venting and avoidance, and significantly and negatively related 18 to distress through these variables. These effects are notable because emotion venting and 19 avoidance are typically involved in mediating effects of illness representations on 20 maladaptive outcomes. These specific indirect effects illustrate that, in some cases, illness 21 representations and the adoption of emotion-focused coping strategies may lead to functional 22 improvements and adaptive outcomes. 23 The specific indirect effects for timeline on outcomes tended to be consistent with the 24 total indirect effects. There were consistent, albeit small, statistically significant and positive specific indirect effects of chronic timeline on physical functioning, social functioning, and 1 well-being mediated by problem-focused generic coping, and significant negative effects on 2 disease state and distress through problem-focused generic coping. Mediation proportion 3 statistics also indicated that effects through problem-focused generic coping accounted for 4 the largest proportion of the total indirect effect of timeline on all outcomes except role 5 functioning. It is important to note that this pattern of effects was inconsistent with the zero-6 order correlations between timeline and outcomes. Timeline denotes chronicity and has been 7 associated with poorer functioning and well-being, and elevated distress in a previous meta- representations tend to be related to maladaptive outcomes such as poorer functioning, greater psychological distress, and increased disease state through adoption of emotion-1 focused coping strategies and lower engagement in problem-focused coping strategies. 2 Sensitivity analysis for sample size. Results of our sensitivity analyses in which we 3 estimated our proposed model for each outcome variable using smaller sample sizes (n = 4 3,130 and n = 1,000) are presented in Tables 11 to 16 (Appendix M, supplemental materials). 5 As expected, 95% confidence intervals about the parameter estimates were progressively 6 wider with decreasing sample size. This meant that the confidence intervals for the smaller 7 parameter estimates were more likely to encompass zero as a possible value. The attenuation 8 effect notwithstanding, non-trivial parameter estimates remained statistically significant 9 according to adopted criteria even in models tested with the smallest sample size. These 10 findings corroborate the imperative of a focus on effect size rather than statistical significance 11 alone when interpreting results from path analyses based on correlations from a meta-12 analysis. Overall, the sensitivity analyses did not alter our interpretation of the pattern of 13 main effects among model constructs. 14

Sensitivity Analyses of Model Effects 15
Although the moderator analysis demonstrated few statistically significant differences 16 in the individual averaged corrected correlations across moderator groups, we conducted 17 sensitivity analyses to test whether the pattern of effects among constructs in our proposed 18 process model was dependent on levels of the study design, illness type, medically-explained 19 vs. medically-unexplained symptoms, and methodological quality moderators 13,14 . Given we 20 13 We did not conduct sensitivity analyses for the time from diagnosis moderator due to large numbers of correlations missing in the matrices for the recently diagnosed moderator group. For example, for correlations among the coping and outcome constructs there was only sufficient data to compute two of the thirty six possible correlations for the recently diagnosed (RD) moderator group. 14 Scores on the methodological quality checklist did not result in substantive variation in the magnitude of the corrected correlations across studies. This was the case for the analyses coded according to checklist score tertiles and a cut-off score of six. Both coding methods revealed extremely similar patterns of effects, although there were many instances where an effect size could not be computed due to insufficient studies for the analysis based on tertiles. We opted to use the coding based on the cut-off scores in subsequent sensitivity analyses to maximize the sample size at each level of the moderator. found that the model including direct and indirect effects of illness representations on 1 outcomes (Model 2) was superior for each outcome variable in full sample analyses, a key 2 purpose of our sensitivity analyses was to test whether this was the case in the moderator 3 groups. We also tested whether the pattern of effects in the proposed models differed across 4 moderator groups by conducting a set of multi-group path analyses of the models 5 constraining each path to be invariant across levels of the moderator using a set of equality 6 constraints. Models were estimated using the averaged corrected correlation matrices as input 7 for the path analyses for each moderator group 15  implications of the present analysis in each area of contribution. 20

Sufficiency of the Model 21
The present research is the first to produce a full meta-analytic inter-correlation 22 matrix among the representation, coping, and outcome variables across studies adopting the 23 common-sense model. Not only did this permit us to ascertain the unique effects of representation and coping dimensions on illness outcomes in multivariate analyses, but to 1 also test the proposed process model based on cumulative evidence from multiple studies. 2 Examining the unique effects in the current analysis, a key finding was that the 3 process model that assumed coping fully mediated effects of illness representations on 4 outcomes was not sufficient, and direct representation-outcome relations were present. This 5 suggests that the coping procedures did not fully explain the effects of illness representations 6 on outcomes. This finding is unique because previous meta-analyses were not able to test the 7 sufficiency of the model. However, given that we found total indirect effects of the illness 8 representation dimensions constructs on illness outcomes through coping means that the 9 mediation hypothesis should not be rejected; instead both direct and indirect effects exist. Our 10 findings suggest that coping partially accounts for the effects of cognitive and emotional 11 representation dimensions on outcomes in chronic illness, but the representations also have 12 unique effects on outcomes independent of the coping constructs included in these studies. 13 The total effects from the models enabled identification of the representation 14 dimensions that, overall, contribute most to explaining variance in illness outcomes. The 15 consequences and identity dimensions emerged as consistent positive predictors of 16 maladaptive outcomes, that is, outcomes related to increased illness progression, greater 17 distress, and poorer well-being, and perceived control as a positive predictor of adaptive 18 outcomes such as better functioning and well-being, and reduced distress and disease 19 progression. Individuals interpreting their illness as having less impact on their life, 20 attributing fewer symptoms to the illness, and perceiving the illness as under control and 21 treatable are likely to experience less distress, better functioning and well-being, and reduced 22 disease state. In addition, emotional representations had a strong positive total effect on 23 distress, a strong negative effect on well-being, and weaker negative effects on physical and 24 role functioning. Individuals who are able to downplay their emotional response to the illness are less likely to experience deleterious emotional outcomes and more likely to report better 1 psychological well-being. By comparison, effects of emotional representations on functioning 2 and illness progression were weaker, suggesting that reduced emotional representations may 3 not have a strong effect on outcomes related to adaptive function and illness recovery. These 4 findings illustrate the overall unique effects of the cognitive and emotional illness 5 representation dimensions on outcomes based on the current sample of studies. 6 One explanation for the insufficiency of the full mediation model is that the coping 7 measures in the included studies may not have been optimally effective in testing mediation. 8 Many studies in the present sample adopted coping measures that tap generalized coping 9 procedures rather than illness-or behavior-specific measures that precisely capture means to well-being, better functioning). In contrast, the total indirect effects from our analysis revealed null or relatively small effects of this construct on outcomes. Furthermore, 1 examination of the specific indirect effects revealed patterns of effect for illness 2 representation dimensions on outcomes that could not be ascertained from the total indirect 3 effects alone. Prominent among these were the specific indirect effects for the perceived 4 consequences dimension which had consistent negative effects on adaptive outcomes through 5 avoidance, and positive effects on adaptive outcomes through problem-focused coping. The 6 presence of both positive and negative specific indirect effects of approximately equal 7 magnitude equated to a zero total indirect effect for consequences on all but one of the illness 8 outcomes in the current analysis. Evaluating the effects of consequences on outcomes 9 through coping based on the total indirect effects alone would lead to an erroneous 10 conclusion that representing the illness as having serious consequences has no indirect effect 11 on outcomes. The specific effects reveal otherwise and indicate that illness consequences can 12 lead to individuals selecting coping strategies that lead to both adaptive and maladaptive 13 outcomes. 14 The presence of effects of illness representation dimensions on both adaptive and 15 maladaptive outcomes mediated by coping has important ramifications for theory. These 16 findings suggest a pattern of effects among constructs in the common-sense model that is 17 more complex than that found in previous research syntheses. Much of the research on the 18 common-sense model has consistently demonstrated that beliefs indicating increased threat, 19 i.e. viewing an illness as having serious consequences, highly symptomatic, and chronic, will 20 lead to emotion-focused coping strategies and poorer outcomes including greater disease 21 progression, lower functioning and well-being, and greater distress (e.g., Dempster et al., exclusively derived from zero-order correlations among constructs and regression models 24 examining effects of representation dimensions on outcomes in the absence of other representation dimensions and coping strategies. The current analysis indicates that these 1 overall patterns, and corresponding conclusions, may be misleading because bivariate 2 analyses do not test the multiple pathways by which representation dimensions relate to 3 outcomes through coping strategies. 4 We have demonstrated that the overall effects of representations on outcomes 5 comprise sets of specific indirect effects that are opposite in sign (i.e., positive and negative). 6 These patterns are, in fact, consistent with theory, but have seldom been shown empirically. determine the specific pattern of effects is, therefore, critical for the accurate prediction of 24 coping responses and concomitant outcomes.

Role of Context and Moderators in the Process Model 1
High levels of heterogeneity were observed in the majority of the effect sizes in the 2 current analysis. This means that after correcting for methodological artifacts there was still 3 substantive variation in the size of the relations among the common-sense model constructs 4 across studies. Given that these effect sizes were used as input for our meta-analytic path 5 analysis of the process model, results must be interpreted in light of the potential of the 6 coefficients involved in the analysis to vary and for that variability to affect the strength of 7 the effects in the process model. The effects reported in the process model, to some extent, 8 reflect a generalized, 'ideal' case of relations in chronic illness, which may be indicative of 9 potential pathways that may operate in the model, but the pathways would be dependent on 10 extraneous moderating variables that determine whether the pathway will be present or 11 absent. A search for moderators was, therefore, warranted. 12 We contended, consistent with theory and previous research on the common-sense It is important to note that our moderator analyses were limited due to substantial 4 heterogeneity remaining in the effect sizes within moderator groups, as well as the small 5 sample sizes in many of the moderator groups. For example, studies on cancer and CVD 6 included a number of different variants of the illness that likely introduced additional within-7 group heterogeneity. More primary research is needed that systematically tests the proposed 8 mediation effects in the process model in the presence and absence of the moderators. 9 Resolving the unique moderating effects of these characteristics may require systematic beliefs about illnesses affect coping strategies and outcomes in chronic disease. 10

Implications for Practice 11
The identification of specific indirect effects in the current analysis has implications 12 for the application of the common sense model in practice. Behavioral interventions should 13 not only target change in certain illness representation dimensions linked to adaptive 14 outcomes (e.g., halting or reversing disease progression, improving functioning, promoting 15 well-being, and allaying distress), but also target the coping strategies that these 16 representations activate. In some cases, targeting change in a given representation dimension 17 would be an appropriate strategy if the dimension was consistently related to adaptive 18 outcomes through the model pathways. For example, the control representation dimension 19 has consistent effects, both direct and indirect, on adaptive illness outcomes. Promoting 20 positive change in this dimension, therefore, would likely lead to adaptive outcomes and it 21 should be labelled as a priority target for intervention. However, in some cases targeting 22 intervention efforts on a particular representation dimension may have effects on adaptive 23 and maladaptive outcomes. For example, providing messages that highlight the serious 24 consequences of an illness to patients may lead to adaptive outcomes by prompting adoption of problem-focused coping strategies, but may, in turn, lead to maladaptive outcomes like 1 distress and disease progression through avoidance. A solution might be to adopt strategies 2 that link the representation with the desired coping strategy. The common-sense model 3 implies that coping strategies may be stored schematically alongside representation characteristics (e.g., illness type and severity, whether the illness is symptomatic or 2 asymptomatic, and whether the illness has medically-explained or medically-unexplained 3 symptoms), personality and individual differences (e.g., optimism, perfectionism), and 4 emotional representations. The proposed moderating effects are illustrated in Figure 3, which 5 outlines the generalized version of the revised model. The basic mediation effects of the 6 process model tested in the current meta-analysis are depicted in the central section of Figure  7 3. In the revised model, these mediated pathways are proposed to be moderated by three sets 8 of factors, represented in the diagram by the effects of illness type, dispositions, and 9 emotional representations on the mediated pathways (see broken lines in Figure 3). In its 10 generalized form, the model indicates the potential for upward and downward moderating 11 effects on the mediated pathways involving representation dimensions, coping strategies, and 12 outcomes (Figure 3). The revised model extends previous theory by providing a formal 13 operationalization of Leventhal et al.'s (1992; proposal that socio-cultural and self-14 system constructs will impact on relations between illness representations, coping strategies, 15 and outcomes (see upper section of Figure 1). 16 We provide specific examples of the moderation effects and illustrate them in Figures  17 4 and 5 16 . The moderating effect of an illness characteristic, the extent to which the illness is 18 treatable, on model pathways is presented in Figure 4. Illnesses and conditions that are 19 unlikely to respond to treatment (e.g., chronic pain) may mean that problem-focused coping 20 efforts will be viewed as less effective, while emotional focused-coping such as venting or 21 even avoidance may be more appropriate. Lower illness treatability may, therefore, moderate 22 the effects of threat and control representations on outcomes (e.g., functioning) through 23 problem-focused coping downwards, and effects of threat representations on outcomes (e.g., 1 distress, disease state) through emotion-focused coping upwards. 2 The moderating effect of emotional representations on model relations is presented in 3 Empirical verification of these pathways should be considered a priority for future research. 18 In addition to incorporating moderating factors, our revised common-sense process 19 model also incorporates the independent effects of beliefs with respect to engaging in specific 20 coping behaviors and beliefs about treatment. The inclusion of these beliefs follows theory 21 and research that has focused on integrating the common-sense model with other social 22 cognitive approaches to understanding coping and illness outcomes. For example, behavioral 23 coping with an illness can be conceptualized as a function of beliefs about the illness (e.g., 24 threat perceptions, perceived control, and emotional responses), as indicated in the common-sense model. It can also be conceived as a function of beliefs about engaging in the coping 1 behavior itself, such beliefs that the behavior will result in desired outcomes (attitudes), 2 beliefs in personal capacity to perform the behavior (self-efficacy), and beliefs in capacity to 3 cope with difficulties or setbacks in managing the illness (coping self-efficacy), as indicated 4 in social cognitive approaches to behavior (e.g., Bandura, 1977;Fishbein & Ajzen, 2009;5 Schwarzer, 2008). The modifications provide a more comprehensive perspective on the 6 factors that are related to problem-focused coping responses to manage illness. For example, 7 an asthmatic patient's decision to engage in a problem-focused coping procedure to manage 8 attacks (e.g., prophylactic use of an inhaler) is not only likely to be a response to beliefs that 9 an attack is sufficiently serious but controllable, but also beliefs that the inhaler will be 10 effective and that they can use it appropriately. The parallel influence of cognitive illness 11 representations and social cognitive beliefs on illness outcomes through problem-focused 12 coping strategies is illustrated in Figure 3 Leventhal et al. (1992;2016) have also suggested that beliefs about treatment, 17 including its efficacy and perceived side effects, would affect selection of, and adherence to, 18 treatment. Treatment can refer to numerous problem-focused coping behaviors including 19 medication adherence and behaviors aimed at rehabilitation, recovery, and prevention of 20 relapse (e.g., physical activity in patients with cardiovascular disease or osteoarthritis). 21 Beliefs relating to treatment behaviors engagement can be accounted for in tests of the 22 common sense model by incorporating constructs from social cognitive theories (e.g., 23 attitudes, self-efficacy). However, there has also been interest in isolating patients' beliefs 24 about medication as a specific form of treatment (Horne, 1997). Horne et al. (1999) contend that medication adherence (e.g., taking anti-hypertension tablets to manage blood pressure) is 1 a function of specific beliefs about the effectiveness (e.g., taking tablets reduces blood 2 pressure at the next test) and drawbacks (e.g., debilitating side-effects of taking the tablets) of A specific example illustrating the effects of additional beliefs in the revised 20 common-sense model is presented in Figure 6. In this model, problem-focused coping 21 procedures are depicted as a response to illness beliefs, consistent with the original model. 22 Coping is also viewed as a response to beliefs about the coping response itself derived from 23 social cognitive theories (e.g., attitudes, self-efficacy, coping self-efficacy). Intentions are 24 also included as a mediator of the effects of illness and behavioral beliefs on coping procedures to reflect the effort and motivation toward the coping response. The relative 1 contribution of each belief set advances theory by illustrating the specific pathways that 2 influence patients' decisions on the adoption of coping procedures. Testing these pathways 3 empirically within the revised model may provide formative evidence to assist in the 4 identification of the beliefs that should be targeted in interventions to promote participation in 5 appropriate coping strategies (Hagger, Hardcastle, et al., 2016;Orbell et al., 2006). The 6 model also provides the opportunity to explore potential interactions between the sets of 7 beliefs. Given research indicating that individuals are more likely to engage in health 8 behaviors in response to a perceived threat when motivation and self-efficacy are high (e.g., 9 Peters, Ruiter, & Kok, 2013), we have indicated that threat perceptions may moderate the 10 effect of intentions on problem-focused coping (see broken line in Figure 6). 11 We also incorporate action planning as an important component of the process 12 preceding the adoption of coping procedures. Leventhal et al. identified action plans as 13 critical to the implementation of coping strategies to manage illness outcomes. The plans 14 identify the specific coping response (e.g., taking an insulin injection), the context in which it 15 will be performed (e.g., in the morning 15 minutes before breakfast), and expectations of the 16 outcomes of the response (e.g., appropriately-managed blood glucose levels). In the model, 17 action plans are depicted as generated by representations of the illness threat and beliefs 18 about the behavior. Furthermore, action plans form a 'bridge' between intentions and the 19 enactment of the coping procedure, as illustrated by the mediation of the effect of intentions 20 on coping by action plans in Figure 6. Formation of action plans has been shown to be pivotal 21 for illness management by assisting individuals in the efficient enactment of an appropriate, 22 effective coping response to a threat representation (Leventhal et al., 2016). The concept of 23 action plans has also been adopted and applied as an important intervention technique in the 24 promotion of health behavior in multiple contexts (e.g., Hagger, Luszczynska, et al., 2016; Orbell, Hodgkins, & Sheeran, 1997;Schwarzer, 2008). Action plans provide opportunities to 1 intervene and promote better adherence to health behaviors. Whereas some patients with 2 chronic illnesses form action plans independently, others need assistance from health 3 professionals. Assisting patients in developing appropriate skills to generate their own action 4 plans may provide important means to enhance coping self-regulation. 5 It is important to note that the proposed effects in our revised common-sense process 6 model are speculative, based on a combination of the findings of the current research, theory, 7 and evidence from primary research to provide potential explanations for effects identified in 8 the current analysis. We expect the revised model to provide a starting point for future 9 research examining the processes by which representations impact outcomes in the common-10 sense model and we have provided examples of some key hypotheses that might be tested. 11 Such research will assist in further advancing the common-sense model and the processes 12 involved in coping with chronic illness. 13

Strengths, Limitations and Recommendations 14
The current analysis has numerous strengths. It is the first test of the sufficiency of a representation dimensions on coping strategies and illness outcomes that may contribute to explaining coping selection and outcomes and the selection of potential targets for behavioral 1 interventions. A further strength of the current analysis is our systematic classification and 2 coding of measures of illness representation, coping, and outcome across research adopting 3 the common-sense model in chronic illness. The theory-based classification was also 4 essential to minimize potential shared variance between constructs attributable to conceptual 5 overlap. Building on a previous coding scheme developed by Hagger and Orbell (2003), we 6 developed a set of definitions of constructs and used expert raters to develop sets of 7 independent illness representation, coping, and outcome categories which accounted for all 8 measures adopted in studies eligible for inclusion in the current analysis. We have provided 9 this coding scheme in the online supplemental materials (see Appendixes F, G, and H, 10 supplemental materials) so that researchers may locate future analyses within the current one, 11 and to assist researchers in categorizing coping and outcome measures. 12 A key limitation of the present analysis is that the vast majority of included studies 13 were correlational in design, those adopting experimental or intervention designs numbered 14 relatively few by comparison. This has also been noted in previous reviews on research Given the heavy dependence on correlational data, causation in the process model is 24 inferred from theory rather than data. While analytic methods such as the path analyses used in the present study imply directional relations, alternative models which specify other 1 directional relations among the variables could be estimated and would be plausible from a 2 statistical and empirical perspective, even if they were not consistent with theory. The 3 correlational nature of the data may account for some of the patterns of effects among model 4 variables identified in the current analysis and in previous research. For example, similar to 5 other meta-analyses, we found negative effects of perceived consequences on problem-6 focused coping and adaptive outcomes, when a theory-based expectation was that such 7 beliefs would serve as a stimulus for problem-focused action focused on treatment to manage 8 the threat. Given the correlational data on which this finding is based, one interpretation 9 would be that the beliefs about consequences is a result of adherence such that success with 10 problem-focused-coping behaviors may have led to better illness outcome and hence fewer 11 perceived consequences. The correlational data does not account for such dynamic processes Overall, we rejected a full mediation model in favor of a model that included both direct and 11 indirect effects of representation dimensions on illness outcomes through coping strategies. 12 We also identified specific mediated pathways which demonstrated that illness 13 representations that signal a health threat, such as perceived consequences, were related to 14 both adaptive and maladaptive outcomes through specific coping strategies. Identification of 15 these specific pathways is important for a full understanding of model effects and conclusions 16 based on zero-order relations or overall pathways could be misleading. Our tests of effects of 17 key moderators revealed few moderation effects and did not resolve the heterogeneity 18 identified in the effect sizes across studies on the model. Many of the studies in the current 19 analysis adopted self-report measures and correlational designs, and we call for research 20 adopting stronger designs, particularly intervention and experimental research, and research 21 using objective measures of specific, behavioral, problem-focused coping strategies and 22 illness outcomes. The research would be extremely informative in resolving some of the 23 relatively untested processes in the common-sense model, such as the dynamic process by 24 which patients' lay representations of illness relate to coping strategy selection, and, subsequently, coping appraisals. We have also proposed a revised common-sense process 1 model that we hope will catalyze primary research testing the effects of moderators, beliefs 2 about coping, and treatment beliefs on coping behavior selection and illness outcomes. We 3 expect findings from the current analysis and revised model will stimulate future research and 4 theory development to advance knowledge on the processes by which illness beliefs affect 5 coping and outcomes in chronic disease.    Figure 3. Path diagram of generalized effects in the revised common-sense process model. Solid lines represent hypothesized effects of beliefs on coping strategies and effects of coping strategies on illness outcomes, and broken lines represent moderating effects. Direct effects of cognitive and emotional representations on illness outcomes omitted for clarity.

Problem-Focused Coping
Cognitive Representations (Threat) Identity Consequences Timeline

Cognitive Representations
Control Coherence

Illness characteristics
Illness type Symptomatic  Figure 4. Path diagram of moderating effects of illness treatability (higher scores represent greater treatability) on mediated relations of perceived consequences and control on illness outcomes mediated by problem-and emotion-focused coping strategies in the revised commonsense process model. Solid lines represent hypothesized effects of representations on coping strategies and effects of coping strategies on illness outcomes, and broken lines represent moderating effects. Direct effects of cognitive and emotional representations on illness outcomes omitted for clarity.  Figure 5. Path diagram of moderating effects of emotional representations (higher scores represent greater emotional responses) on mediated relations of perceived consequences and control on illness outcomes mediated by problem-and emotion-focused coping strategies in the revised common-sense process model. Solid lines represent hypothesized effects of representations on coping strategies and effects of coping strategies on illness outcomes, and broken lines represent moderating effects. Direct effects of cognitive and emotional representations on illness outcomes omitted for clarity. ) on problem-focused coping behaviors in the revised common-sense process model. Direct effects of cognitive and emotional representations on illness outcomes omitted for clarity. The proposed model is a generalized one with constructs comprising multiple dimensions of constructs (e.g., behavioral beliefs, cognitive representations) that may have effects on outcomes with different signs.