Kaohsiung J Med Sci June 2007 • Vol 23 • No 6 287
Studies on the association between stress and health have been conducted for many years [1–7]. However, the research findings are inconsistent in terms of their significance, magnitude, and, in some cases, direction. Some studies found a significant positive relationship between stress and health distress [8–11]. Some studies
showed little association between them. Other studies even demonstrated a negative correlation between them [3,12–14].
These inconsistent findings could result from three causes. First, different studies investigated different types of stress. For example, some studies focused on specific work stress or caregiver’s stress, while others focused on the measure of general life stress. Second, different studies were concerned with different health facets. For example, some studies focused on health index, such as clinical diseases or symptoms, others focused on social role functions or adaptive behavior,
Received: August 18, 2006 Accepted: November 10, 2006 Address correspondence and reprint requests to: Chao-Hung Chiu, Clinical Psychologist, Students Counseling Center, Far East University, 49 Chung-Hua Road, Hsin-Shih, Tainan 744, Taiwan. E-mail: emmanue@ms23.hinet.net
A M
ETA
-
ANALYSIS OF THE
A
SSOCIATION
B
ETWEEN
S
TRESS AND
H
EALTH IN
T
AIWAN
Lifa Yu,1Yaw-Sheng Lin,2Jew-Wu Chen,1Hsiu-Hung Wang,3and Chao-Hung Chiu4
1Faculty of Psychology, College of Health Science and 3Faculty of Nursing, College of Nursing, Kaohsiung
Medical University, Kaohsiung, 2Department of Clinical and Counseling Psychology, National Dong Hwa University, and 4Students Counseling Center, Far East University, Tainan, Taiwan.
This study adopted the meta-analysis technique to analyze 354 journal articles, theses, and disser-tations that had investigated the association between stress and health in Taiwan between January 1980 and December 2003. This study was conducted with the purpose of understanding the association between general stress and general health, the discrepant associations between different stress types and health facets, and the possible moderators between general stress and general health. A computer search for relevant studies was conducted on several databases using the key words “stress” and “life event”. For each eligible study, the important study characteris-tics were recorded, and the effect sizes of the relationship between stress and health were com-puted. Furthermore, in order to investigate the moderating effects of the study characteristics on the stress–health relationship, the methods of categorical model analysis and correlation analysis were employed. The results of this study revealed that: (1) the correlations between general stress and general health as well as between general stress and various health facets fell between medium and high; (2) there existed different degrees of association between various stress types and health facets; and (3) none of the demographic and methodologic variables could by itself moderate the relationship between general stress and general health as the moderator effects were not sufficient and strong enough. This study presents a multidimensional framework of the issues on the relationship between stress and health, and it provides guiding references for future research. No evidence was found for moderating effects of social support, coping strategies, and personality traits on the stress–health relationship. Such findings may be due to methodologic limitations. This suggests that further investigation is needed.
Key Words:health, meta-analysis, stress (Kaohsiung J Med Sci 2007;23:287–97)
and still others focused on subjective life quality. Third, some moderators may exist between stress and health—such as social support, coping strategies, per-sonality traits, demographic variables, study quality, and so on—which alter the association between stress and health [15–17]. In sum, “stress” and “health” are multidimensional concepts [18–19]. Different stress types and health facets could result in different degrees of association, as shown in many studies. The explora-tion of moderators between stress and health is draw-ing more attention. A moderator is a qualitative or quantitative variable that affects the direction and/or strength of the relationship between an independent variable and a dependent variable.
With a view to achieving better understanding of the association between general stress and general health, the discrepant associations between different stress types and health facets, and the possible mod-erators between general stress and general health, this study adopted the meta-analysis method to system-atically re-analyze the findings of related studies on stress and health in Taiwan. Meta-analysis, a quantita-tive method of summarizing existing studies, is defined as an analysis of analyses. That is, the pooled results of individual studies that have previously appeared to be contradictory or confusing are re-analyzed to provide a systematic, quantitative review of the data, and thus to arrive at strong, credible conclusions.
M
ATERIALS ANDM
ETHODSConceptualization and classification of
stress and health
Stress is a multidimensional concept. In the theoreti-cal dimension, stress has been viewed as a stimulus, a response, or a process. In short, the stress defined by the stimulus approach is the stressors, or objective stressful events; the stress defined by the response approach is the strains, or one’s reactions to stressors. The stress defined by the process approach puts em-phasis on one’s subjective appraisal of the demands of environments [16]. On the basis of the investigation of stress and health constructs, we placed the concepts of the response approach in the health domain and the concepts of stimulus and process approaches in the stress domain in this study.
In the research field classification dimension, the trend in recent studies has been to classify studies into
different research fields according to the contexts in which different life events occur. Such research fields include, for example, work stress [4], stress of care-giving [5], illness stress [6], stress of military service [7], etc. Thus, after reviewing the related studies in our meta-analysis and following the classification of recent research fields, we divided the general stress into seven stress types: stresses of caregiving, military service, major disaster, work, practicum, illness, and student life.
Health is also a multidimensional concept. In this study, we adopted and expanded the concept of health on the basis of Smith’s framework [20]. Accord-ing to Smith, the concept of health involves a grada-tion of health or illness, which can be divided into four distinct facets: clinical, adaptive, role-performance, and eudemonistic. First, the clinical facet relates to the presence or absence of signs or symptoms of disease or disability in the individual as identified by med-ical science. In this study, we further subdivided the clinical facet into physical and psychologic facets. The former included variables such as stomach ache, hypertension, etc., and the latter included depression, anxiety, etc. Second, the adaptive facet, including such variables as health behavior, social behavior, self-esteem, etc., is concerned with the extent to which the individual maintains flexible adaptation to the envi-ronment. Third, the role-performance facet, includ-ing variables such as job involvement, organizational commitment, etc., relates to the degree to which the individual performs social roles with expected out-put. Fourth, the eudemonistic facet, including such variables as job satisfaction, life satisfaction, quality of life, etc., is concerned with the degree of exuberant wellbeing and satisfaction. In this meta-analysis, these above-mentioned health facets were included in the concept of general health.
Literature search
This quantitative review included studies conducted between January 1980 and December 2003 in Taiwan. A computer search of the following databases was conducted: PerioPath—Index to Chinese Periodical Literature, Dissertation and Thesis Abstracts System, and NSC Science and Technology Information System. The key words used to identify studies included “stress” and “life event”. The search was restricted to studies published in Chinese and involving human subjects. The references cited in the studies identified
by the above approach were also used to locate addi-tional studies.
Three inclusion criteria were adopted in this meta-analysis. First, the studies must examine the relation-ship between stress and health. Second, they must provide information sufficient for the computation of effect sizes. Third, for duplicate studies, only those that provided the most comprehensive and necessary information were selected. In the preliminary liter-ature search, the abstracts of 722 dissertations and theses and 746 research articles were identified and reviewed. Of these studies, 354 studies, including 281 dissertations and theses and 73 research articles, met the inclusion criteria and were included in our meta-analysis. A list of these 354 studies is available from the corresponding author upon request.
Study characteristics coded from
each study
A coding sheet was designed to record the three types of study characteristics from each study: demographic, methodologic, and substantive. The first two types included the study characteristics of predominant sex, age, marital status, education level, occupation, socio-economic status, date of publication, type of publica-tion, and study quality. The third type included the study characteristics of social support, coping strate-gies (including problem-focused coping and emotion-focused coping), and personality traits (including Type A/Type B personality and internal/external locus of control type of personality).
Based on the methodologic suggestions outlined by Brown [21], and on our knowledge obtained from some literature reviews of stress and health, the crite-ria for evaluating the quality of a study included the following items: sampling method (e.g. convenience sample, random sample; maximum 15 points), sample size (e.g. ≤ 30, 31–100, 101–500; maximum 15 points), specification of study sample (e.g. incomplete descrip-tion, complete description; maximum 10 points), valid-ity of instruments (e.g. describing validvalid-ity of less than half of the instruments, describing validity of more than half of the instruments; maximum 10 points), reliability of instruments (e.g. all scales α ≥ 0.90 or test–retest reliability/split-half reliability/subscales α ≥ 0.80, all scales α ≥ 0.80 or test–retest reliability/ split-half reliability/subscales α ≥ 0.70; maximum 20 points), and appropriateness of instruments for meas-uring alleged concepts (e.g. somewhat appropriate,
very appropriate; maximum 20 points). We then summed all item scores for each study and ranked all the studies from the highest (90 points) to the lowest (34 points) according to their total scores and divided them into three groups—low, medium, and high quality studies—based on an equal interval.
As far as the substantive study characteristics were concerned, because many original studies used different instruments to measure the same construct (e.g. social support), we converted these different measures for the same construct to a common metric scale, with scores ranging from 0 to 100, to make them comparable. Different stress types and health facets were also categorized and recorded such that the higher the score on health outcome measures, the poorer the health conditions.
Interrater agreement
Forty-eight studies were randomly drawn from the 354 studies and independently coded by two coders. Both coders are licensed clinical psychologists. The interrater agreement ranged from 96% to 100% for the demographic, methodologic and substantive study characteristics, and 95–100% for the categorization of health facets and stress types. Disagreements in coding were eventually resolved through discussion.
Computation and analysis of effect sizes
Each study result was represented in the form of effect sizes. The effect size estimate used in this meta-analysis was r, the correlation between stress and health in each original study. We used Hedges and Olkin’s [22] meta-analysis method to determine the average (or mean) effect sizes for the relationships between general stress and general health, between general stress and various health facets, and between various stress types and health facets across all studies.First, we transformed all effect size estimates by using Fisher’s Z transformation to reduce the effects of non-normality of the sampling distribution. When more than one effect size was available from a single sample (i.e. a single study), we averaged these effect sizes. To correct for sampling error, we weighted each Fisher’s Z transformed correlation by its sample size. Second, we calculated the mean weighted Fisher’s Z transformed correlation for each stress–health rela-tionship mentioned above. Furthermore, with regard to general stress and general health, we examined the variations among effect sizes through a QT test of
homogeneity to determine whether the effect size estimates were relatively consistent. In the absence of homogeneity, the study characteristics were used to account for variability in heterogeneous effect sizes. For the categorical demographic and methodologic study characteristics, we performed tests of categorical models to determine the relationship between these study characteristics and the magnitude of the effect sizes. For each study characteristic, the categorical model calculated a between-class homogeneity statis-tic (QB) and several within-class homogeneity statistics (QW). A significant QBvalue suggested that the effect
size estimates differed across classes (subgroups) of an identified study characteristic and that the study characteristic identified might be an important mod-erator of effect size estimates, provided that the ef-fect size estimates within classes were found to be homogeneous (i.e. QWstatistics were not significant).
Significant QWvalues, on the other hand, suggested that the study characteristic was not a strong, suffi-cient moderator because effect sizes remained hetero-geneous within classes. In the computational process of categorical model analysis, the mean weighted Fisher’s Z transformed correlation of each class was also obtained.
Finally, we transformed all the mean weighted Fisher’s Z transformed correlations back to the original correlation metric scale for the purpose of easy inter-pretation. The Comprehensive Meta-analysis program (Biostat Inc., Englewood, NJ, USA) [23] was used for data analysis.
Owing to the continuous nature of three substantive study characteristics: social support, coping strategies, and personality traits, we examined bivariate correla-tions between each of the three study characteristics
and stress–health effect sizes to identify the study characteristics that might explain substantial portions of effect size variance and further act as the moderators.
R
ESULTSAs shown in Table 1, across the 354 studies aggre-gated, the overall mean weighted effect size (r+ or mean weighted correlation) was 0.359 (p< 0.001, 95% confidence interval [CI]= 0.355/0.364), significantly different from zero, indicating a positive association between general stress and general health distress. However, calculation of the homogeneity QTstatistic
(QT= 7,061.141, p < 0.001) indicated significant hetero-geneity among effect sizes. Therefore, study character-istics were used to account for variability in the effect sizes. The tests of categorical models were performed for the demographic and methodologic study charac-teristics, and the results are presented in Table 3 in this section. Also, Table 1 shows that in terms of the magni-tude of associations between general stress and vari-ous health facets, the eudemonistic health facet was the highest (r+= 0.381) and the adaptive health facet was the lowest (r+=0.281), with the clinical health facets falling in between. All of the mean weighted effect sizes presented in Table 1 were significantly different from zero in terms of 95% CI, indicating positive relation-ships between general stress and distress of various health facets.
Regarding the different combinations of stress types and health facets, the mean weighted effect sizes of all the combinations were significantly different from zero, ranging from 0.165 (p<0.001) for the combi-nation of disaster stress and clinical–psychologic health
Table 1.The mean weighted effect sizes of the relationship between general stress and general health as well as between general stress and various health facets
Health facet k r+ 95% CI for r+
General 354 0.359* 0.355/0.364 Clinical 185 0.356* 0.350/0.362 Clinical–physical 88 0.308* 0.299/0.317 Clinical–psychologic 127 0.367* 0.360/0.374 Role-performance 123 0.303* 0.295/0.311 Adaptive 96 0.281* 0.273/0.289 Eudemonistic 96 0.381* 0.373/0.390
*p< 0.001. k = number of studies/effect sizes in the health facets; r+= mean weighted effect size (i.e. mean weighted correlation); CI =
facet to 0.485 (p<0.001) for that of caregiver’s stress and clinical–psychologic health facet. This suggested that there existed different degrees of positive association between different stress types and distress of various health facets. Table 2 shows the mean weighted effect sizes of different health facets in each stress type. In terms of the largest effect sizes of health facets in each stress type, almost all the stress types brought the greatest impacts on the clinical–psychologic health facet and eudemonistic health facet. Among them, care-givers’ stress was the type of stress that had the great-est influence on health. Specifically, the association
between stress of caregivers and clinical–psychologic health facet was the largest, whereas the relationship between the stress of student life and the same psy-chologic facet was the smallest, and the association between work stress and the same psychologic facet fell in between.
Table 3 presents tests of categorical models by demographic and methodologic study characteris-tics. Almost all of the mean weighted effect sizes (r+) within each class were above 0.30, significantly differ-ent from zero with 95% CIs, which indicated positive relationships between general stress and general health
Table 2.The mean weighted effect sizes of different health facets in each stress type
Stress type Health facet k r+ 95% CI for r+
Caregiving Clinical–psychologic 10 0.485* 0.427/0.540
Clinical–physical 7 0.464* 0.394/0.529
Clinical 15 0.387* 0.342/0.431
Adaptive 6 0.319* 0.242/0.393
Eudemonistic 5 0.241* 0.149/0.329
Military service Eudemonistic 3 0.473* 0.437/0.509
Clinical 5 0.378* 0.347/0.407
Adaptive 3 0.354* 0.326/0.380
Clinical–psychologic 3 0.273* 0.225/0.319
Clinical–physical 3 0.268* 0.220/0.314
Role-performance 2 0.212* 0.140/0.282
Major disaster Clinical–physical 1 0.410* 0.283/0.523
Adaptive 1 0.400* 0.272/0.514 Clinical 4 0.202* 0.161/0.243 Clinical–psychologic 3 0.165* 0.117/0.213 Work Clinical–psychologic 51 0.399* 0.387/0.410 Clinical 68 0.379* 0.368/0.389 Eudemonistic 74 0.350* 0.340/0.361 Clinical–physical 43 0.319* 0.306/0.331 Role-performance 113 0.315* 0.306/0.323 Adaptive 34 0.254* 0.239/0.269 Practicum Clinical–psychologic 4 0.395* 0.357/0.431 Clinical 9 0.361* 0.334/0.388 Clinical–physical 3 0.176* 0.108/0.243 Illness Eudemonistic 4 0.387* 0.300/0.467 Clinical–psychologic 9 0.384* 0.333/0.434 Clinical 13 0.368* 0.323/0.412 Clinical–physical 4 0.309* 0.223/0.390 Adaptive 11 0.308* 0.262/0.352 Role-performance 1 0.190* 0.047/0.325
Student life Clinical–psychologic 29 0.327* 0.316/0.339
Clinical 43 0.325* 0.314/0.335
Clinical–physical 16 0.287* 0.269/0.304
Adaptive 27 0.262* 0.249/0.274
Eudemonistic 2 0.204* 0.153/0.254
*p< 0.001. k = number of studies/effect sizes in different health facets of each stress type; r+= mean weighted effect size (i.e. mean
Table 3.Tests of categorical models for effect sizes of the relationship between general stress and general health by demo-graphic and methodologic study characteristics
Variable/class k r+ 95% CI for r+ QB QW Predominant sex 137.471* < 5% female 20 0.407* 0.393/0.421 861.184* 5–44% female 73 0.352* 0.342/0.362 1,108.058* 45–54% female 76 0.337* 0.328/0.345 2,156.459* 55–94% female 94 0.390* 0.381/0.398 1,935.785* ≥ 95% female 75 0.337* 0.326/0.349 635.027* Cannot tell 16 0.350* 0.328/0.372 227.157* Age (yr) 148.835* ≤ 18 49 0.360* 0.351/0.369 1,488.190* 19–30 76 0.314* 0.303/0.324 867.850* 31–45 55 0.408* 0.396/0.419 1,161.472* ≥ 46 25 0.368* 0.350/0.385 183.012* Mixed 149 0.364* 0.357/0.371 3,211.782* Marital status 5.857 Married 146 0.367* 0.360/0.374 2,440.880* Single 96 0.357* 0.348/0.362 2,269.888* Mixed 112 0.355* 0.348/0.365 2,344.517* Education level 198.644* Elementary 13 0.314* 0.295/0.334 150.076* Junior high 26 0.360* 0.349/0.372 1,038.603* Senior high 30 0.350* 0.338/0.362 959.712* Junior college 26 0.460* 0.445/0.475 477.759* University 77 0.338* 0.327/0.348 1,235.207* Mixed 182 0.359* 0.353/0.365 3,001.140* Occupation 316.130* Accountant 6 0.424* 0.381/0.466 103.106* Police 8 0.411* 0.393/0.428 551.569* Nurse 15 0.399* 0.377/0.420 54.677* Teacher 17 0.393* 0.368/0.417 103.940* Soldier 7 0.361* 0.340/0.382 43.778* Student 73 0.342* 0.334/0.350 1,945.228* Factory worker 6 0.202* 0.153/0.251 61.430* Engineer 5 0.171* 0.129/0.212 34.629* None 41 0.311* 0.299/0.323 913.091* Mixed 176 0.384* 0.377/0.391 2,933.563* Socioeconomic status 60.492* High 42 0.395* 0.382/0.409 755.597* Medium 32 0.388* 0.374/0.403 711.696* Low 85 0.343* 0.336/0.351 2,349.733* Mixed 195 0.358* 0.352/0.365 3,123.623* Date of publication 107.848* 2001–2003 78 0.401* 0.392/0.410 2,176.390* 1991–2000 200 0.352* 0.346/0.358 3,560.070* 1980–1990 76 0.337* 0.337/0.346 1,216.833* Type of publication 40.737*
Dissertation and thesis 281 0.367* 0.362/0.372 6,076.044*
Journal article 73 0.332* 0.323/0.342 944.360*
Study quality 211.989*
High 88 0.395* 0.389/0.402 3,113.316*
Medium 201 0.344* 0.338/0.351 2,955.909*
Low 65 0.305* 0.295/0.317 729.926*
*p< 0.001. k = number of studies/effect sizes in the class or subcategory; r+= mean weighted correlation; CI = confidence interval;
distress. As far as the highest associations between general stress and general health were concerned, they were studies with < 5% female subjects for predomi-nant sex variable, studies with 31–45-year-old subjects, studies with married subjects for marital status, stud-ies with junior college subjects for education level, studies with accountant subjects for occupation, and studies with subjects with high socioeconomic status. Among various occupations, the magnitude of associ-ations obtained from studies with accountants, police, nurses, and teachers as subjects was close and high. Moreover, the impact of stress on health increased in recent studies. The associations obtained from theses and dissertations were higher than those obtained from journal articles, and the higher the quality of a study, the greater the association.
As shown in Table 3, the between-class hetero-geneity (QB) was significant for all of the tested
cate-gorical models except for marital status. Marital status yielded a QBvalue that was also very close to the 0.05
significance level (p= 0.054). However, the Qwstatistics
were significant at the 0.001 level in all of the classes. Therefore, although these study characteristics had some moderating effects because of their significant between-class effect size differences, none of these study characteristics could be regarded as strong, suf-ficient moderators because the effect sizes remained significantly heterogeneous within each class.
As for the substantive study characteristics, all of the correlations between each of the study character-istics (social support, coping strategies, personality traits) and effect sizes obtained from the relationship between general stress and general health were not significant, ranging from −0.141 (p = 0.449) between internal/external control type of personality and stress–health effect sizes to 0.215 (p= 0.093) between problem-focused coping and stress–health effect sizes. This indicated that these three substantive study char-acteristics were not moderators in the stress–health relationship.
D
ISCUSSIONOwing to the influence of many factors, the answer to the question “How much is stress and health associ-ated?” was still inconclusive [2,24–26]. After systemat-ically meta-analyzing a large body of original studies, the current study revealed that the overall association
between general stress and general health reached 0.359 (p< 0.001), indicating a positive relationship between stress and health distress. According to Cohen’s guide-lines [27] for small (r≤ 0.10), medium (r = 0.25) and large (r≥ 0.40) effects, the association of 0.359 was far above the medium level.
After comparing the results of this study with those of other meta-analyses [24,28–30], it was found that the rank of magnitudes of association between gen-eral stress and different health facets was similar. The facets associated with general stress, listed from large to small magnitude, were the eudemonistic facet, clinical–psychologic facet, clinical–physical facet, role-performance facet, and adaptive facet.
Moreover, this study also found that the eude-monistic facet and clinical–psychologic facet were closer in magnitude of association, and the clinical– physical facet, role-performance facet, and adaptive facet were closer but lower in strength of association. This seems to show two clusters. There are three possi-ble explanations for this lower cluster. First, according to the three stages of the general adaptation syndrome theory proposed by Selye [31], we may infer that the time sequence pattern of association between stress and clinical–physical health facet would be relatively high at first, becoming lower, and finally returning to high again. However, this study did not take the time variable into account. Owing to coding diffi-culty, we did not recode the time period in which the health outcomes were measured as a study charac-teristic. Therefore, it was impossible to examine the change of the influence that the stress imposed on clinical–physical facet over time. Simply aggregating data obtained from different time points might result in central tendency effects. Second, Yerkes and Dodson [32] pointed out that the relationship between work stress and work performance presented an inverted U-shape. This made possible the further inference that the relationship between stress and health might not be linear. This inference was also supported by Chen [33]. In our meta-analysis, work performance was catego-rized as an important variable of role-performance health facet, and thereby the relationship between stress and role-performance facet might become lower. Third, in terms of sensitivity to the stress change, compared with other health facets, the adaptive health facet was relatively not sensitive. For example, when faced with stress, one may soon show clinical symp-toms, such as nervousness, anxiety, or stomach ache.
In terms of the role-performance facet, one may work efficiently soon after being exposed to stress and then may become tired later. In terms of the eudemonistic facet, when encountering stress, one may intuitively feel or experience poor life quality and satisfaction. In contrast, in the adaptive facet, the behaviors such as health behavior, social behavior, deviated behavior, self-esteem, etc., are long-term and stable behavior patterns and mental status, which may not change concurrently or instantly along with stress change.
Because different stress types would impose threats on different health facets, this study picked out the most influential stress type, caregiving, and found that clinical–psychologic and clinical–physical facets of caregivers’ health were under the greatest threat. The related literature reviews conducted by Chiu et al [34] concluded that the greatest stresses the caregivers faced were their restricted work schedules, lack of social support, the impacts their families faced, inade-quate attendance knowledge, and worry over the sick; the physical health problems they suffered included tiredness and poor quality of sleep; and the psychologic health problems they experienced were frustration, anxiety, and hopelessness. Caregivers’ clinical health problems deserve our serious concern. Although the eudemonistic facet of military servicemen’s health was also under a great threat, the results of a very small number of studies (five studies or less) in this category are not stable or representative [35].
After a series of examinations in this study, Aneshensel’s viewpoint [17] that the demographic variables have moderating effects on the stress–health relationship was not strongly supported. Nearly all of the between-class effect size differences were sig-nificant, which implied that these variables had some moderating effects. However, the studies’ effect sizes remained heterogeneous in all of the classes or sub-groups. That is to say, none of these demographic study characteristics could by itself moderate the association between stress and health. We anticipate that the moderating effects of these study characteris-tics can be significant if these study characterischaracteris-tics interact with each other. We also anticipate that when the heterogeneous classes of a study characteristic are further subdivided on the basis of a second study char-acteristic, the heterogeneity within subclasses may still remain, even if we conduct higher-order categorical model analyses. This means that many more study characteristics may be needed in combination for
moderating the association between stress and health more completely and adequately. This finding about the moderating effects can also be applied to the methodologic characteristics, such as publication date, publication type, and study quality, on the stress–health relationship.
With regard to social support, coping strategies, and personality traits, this study did not find that they functioned as moderators between stress and health. This finding is inconsistent with the viewpoints held by scholars in the field of health psychology [15,16]. In this meta-analysis, many original studies used different instruments or different scales to mea-sure the same construct. In order to make these origi-nal scale scores comparable, it was necessary to enlarge the original scale scores and convert them to a common metric system. However, in the process of enlarging and converting scale scores, some data might lose their original meaning. On the other hand, since the numbers and definitions of subscales yielded by the instruments were different in the original studies, we had to recategorize them according to our coding scheme. In so doing, some problems might inevitably result. Therefore, the suggestion for future research is that it would be better to reexamine the appropri-ateness of research methods of conversion and recat-egorization than to boldly claim that social support, coping strategies, and personality traits are not erators between stress and health. Clarifying the mod-erating effects of these study characteristics should remain the direction for future meta-analysis studies on this issue.
As mentioned earlier, the impacts of stress on dif-ferent health facets were difdif-ferent under difdif-ferent sit-uations. This implies that we cannot focus on only one single facet in the course of health promotion. According to Smith [20], there are certain significant differences in outlook and emphasis among the four facets of health. Both the clinical and role-performance facets seem to focus on the maintenance of stability. In contrast, the adaptive and eudemonistic facets are ori-ented toward change and growth. How to strike a bal-ance among these four facets should be the goal.
Based on the review of these original studies, it was found in recent studies that stress imposed a consid-erable threat. This indeed provides a significant warn-ing of health threat and indicates that stress will play a formidable role in health maintenance and health promotion in future.
We have one final suggestion for future research. This current study only focused on investigating the possible moderators that might influence the associa-tion between general stress and general health. Future meta-analyses should focus on investigating the roles that moderator variables play in the relationships between different stress types and health facets.
A
CKNOWLEDGMENTSThe authors thank the National Science Council, Taiwan, for funding this research (NSC 92-2413-H-037-003).
R
EFERENCES1. Aakster CW. Psycho-social stress and health distur-bances. Soc Sci Med 1974;8:77–90.
2. Brown MA. Social support, stress, and health: a compar-ison of expectant mothers and fathers. Nurs Res 1986;35: 72–6.
3. Haines MM, Stansfeld SA, Job RFS, et al. Chronic air-craft noise exposure, stress responses, mental health and cognitive performance in school children. Psychol Med 2001;31:265–77.
4. Hammer TH, Saksvik PO, Nytro K, et al. Expanding the psychosocial work environment: workplace norms and work-family conflict as correlates of stress and health. J Occup Health Psychol 2004;9:83–97.
5. Lee HL, Tang MF, Huang SS. The stress and coping behavior of the family members of cancer patients dur-ing the hospitalization period. Kuo-Fang Med J 1994; 9:485–9.
6. Cheng HY. Relationships between stress, coping strate-gies, and quality of life in adult asthma patients. J Nurs
Res 2000;8:599–613.
7. Sun MH. Stressors, coping, psychological and somatic health among the soldiers. Chin J Ment Health 1995;8: 59–74.
8. Tsay WP, Lee S, Lo SK. Study on the feeling of stress and the psychophysiologic reaction of new recruits.
Chang Gung Nurs 1996;7:57–68.
9. Pan IL, Tien SF, Chang M. Stressors, coping behaviors and physical-mental health status among primary care-givers of bedridden patients at home. Public Health 1998; 24:219–34.
10. Hamner WC, Tosi HL. Relationship of role conflict and role ambiguity to job involvement measures. J Appl
Psychol 1974;59:497–9.
11. Beard F. Marketing client role ambiguity as a source of dissatisfaction in client-ad agency relationships. J Advert
Res 1996;36:9–20.
12. Luo ZM. Life Stress, Adaptive Strategies and Mental Health. Unpublished master’s thesis, Taipei: National Taiwan University, 1983.
13. Lin YS. The Effects of Life Events, Coping Strategies and
Social Support on Psychological Adaptability of the Elderly.
Unpublished master’s thesis, Taipei: National Taiwan University, 1988.
14. Chang CI. The Relationships of Dysmenorrhea, Life Stress
and Free Radical in Adolescence. Unpublished master’s
thesis, Taipei: National Taipei College of Nursing, 2002.
15. Ogden J. Health Psychology: A Textbook. Buckingham, Philadelphia: Open University Press, 1996.
16. Sarafino EP. Health Psychology: Biopsychosocial Interactions, 4thedition. New York: John Wiley & Sons, 2001. 17. Aneshensel CS. Outcomes of the stress process. In:
Horwitz AV, Scheid TL, eds. A Handbook for the Study of
Mental Health: Social Contexts, Theories and Systems, 1st edition. Cambridge, UK: Cambridge University Press, 1999:211–27.
18. Wheaton B. The nature of stressors. In: Horwitz AV, Scheid TL, eds. A Handbook for the Study of Mental
Health: Social Contexts, Theories and Systems, 1st edition. Cambridge, UK: Cambridge University Press, 1999: 176–97.
19. Wang HH, Wu SZ, Liu YY. Association between social support and health outcomes: a meta-analysis. Kaohsiung
J Med Sci 2003;19:345–51.
20. Smith JA. The idea of health: a philosophical inquiry.
Adv Nurs Sci 1981;3:43–50.
21. Brown SA. Measurement of quality of primary studies for meta-analysis. Nurs Res 1991;40:352–5.
22. Hedges LV, Olkin I. Statistical Method for Meta-analysis. Orlando, FL: Academic Press, 1985.
23. Borenstein M, Rothstein H. Comprehensive Meta-analysis:
A Computer Program for Research Synthesis, Version 1.1.
Englewood, NJ: Biostat, 1999.
24. Lee CH, Wu WY, Tsai DC. Relationship between role stress of the employees to their job satisfaction, job anx-iety, and job performance—application of meta-analysis method, J Natl United Univ 2002;19:303–26.
25. Schuler RS. Role perceptions, satisfaction, and perform-ance: a partial reconciliation. J Appl Psychol 1975;60: 683–7.
26. Rabkin JG, Struening EL. Life events, stress and illness.
Science 1976;194:1013–20.
27. Lipsey MW, Wilson DB. Practical Meta-analysis. Thousand Oaks, CA: Sage Publications, 2001.
28. Hwang BY. The Theory Testing of Combing Meta-analysis
and Structural Equations Modeling. For example: Stress-moderator-response Model’s Building and Confirmation.
Un-published doctoral dissertation, Taipei: National Taiwan Normal University, 2004.
29. Yen YN. A Meta-analysis on Related Variables of Teachers’
Burnout. Unpublished master’s thesis, Chia-Yi: National
30. Mills LJ. Evaluation of a Model of Hardiness and Health
Using Meta-analysis and Path Analysis. Unpublished
doctoral dissertation, Tulsa: University of Tulsa, 2000. 31. Selye H. The Stress of Life. New York: McGraw-Hill, 1956. 32. Yerkes RM, Dodson JD. The relation of strength of habit-formation. J Comp Neuro Psychol 1908;18:459–82. 33. Chen HC. Job Stress and Mental Health and Burnout: An
Examination of Person-environment Fit in the Work Domains.
Unpublished master’s thesis, Taipei: National Chengchi University, 2000.
34. Chiu CJ, Hsu SM, Wu CM. Literature review of primary caregivers’ burden, stress and coping in Taiwan research.
J Health Sci 2002;4:273–90.
35. Arthur W, Bennett W, Huffcutt A. Conducting
Meta-analysis Using SAS. Mahwah, NJ: Lawrence Erlbaum
!"VR==U==NU= !"VR==NN==NM= !"#$%&'()* !"##$%&'( !"#$%&QV
