Upload
daniassalam-rohul
View
220
Download
1
Tags:
Embed Size (px)
Citation preview
Effects of Occupational Stress Management Intervention Programs:A Meta-Analysis
Katherine M. Richardson and Hannah R. RothsteinBaruch College, City University of New York
A meta-analysis was conducted to determine the effectiveness of stress management interventionsin occupational settings. Thirty-six experimental studies were included, representing 55 interven-tions. Total sample size was 2,847. Of the participants, 59% were female, mean age was 35.4, andaverage length of intervention was 7.4 weeks. The overall weighted effect size (Cohens d) for allstudies was 0.526 (95% confidence interval ! 0.364, 0.687), a significant medium to large effect.Interventions were coded as cognitivebehavioral, relaxation, organizational, multimodal, oralternative. Analyses based on these subgroups suggested that intervention type played a mod-erating role. Cognitivebehavioral programs consistently produced larger effects than other typesof interventions, but if additional treatment components were added the effect was reduced.Within the sample of studies, relaxation interventions were most frequently used, and organiza-tional interventions continued to be scarce. Effects were based mainly on psychological outcomevariables, as opposed to physiological or organizational measures. The examination of additionalmoderators such as treatment length, outcome variable, and occupation did not reveal significantvariations in effect size by intervention type.
Keywords: stress management, meta-analysis, employee intervention
Employee stress has increasingly become a con-cern for many organizations. To paraphrase the fa-ther of stress, Hans Selye, stress is an unavoidableconsequence of life, and therefore an unavoidableconsequence of organizations. Americans are work-ing longer and harder, and job stress continues toincrease. The average work year for prime-age work-ing couples in the United States increased by nearly700 hours in the past two decades (Murphy & Sauter,2003; U.S. Department of Labor, 1999). From 1997to 2001, the number of workers calling in sick be-cause of stress tripled. The American Institute ofStress reported that stress is a major factor in up to80% of all work-related injuries and 40% of work-place turnovers (Atkinson, 2004). This is not solelyan American phenomenon. The Confederation ofBritish Industry reported stress as the second highestcause of absenteeism among nonmanual workers inthe United Kingdom, and the European Foundationfor the Improvement of Living and Working Condi-tions reported that stress affects a third of the Euro-
pean working population (Giga, Cooper, & Faragher,2003). In Australia, most states report an increasingnumber of annual workers compensation claimsresulting from workplace stress (Caulfield, Chang,Dollard, & Elshaug, 2004). Organizations provide amajor portion of the total stress experienced by aperson as a result of the amount of time spent on thejob, the demands for performance, and the interactionwith others in the workplace (DeFrank & Cooper,1987).
Although it is not possible to eliminate stress en-tirely, people can learn to manage it. Many organi-zations have adopted stress management training pro-grams to try and reduce the stress levels of theirworkforce. A stress management intervention (SMI)is any activity or program initiated by an organizationthat focuses on reducing the presence of work-relatedstressors or on assisting individuals to minimize thenegative outcomes of exposure to these stressors(Ivancevich, Matteson, Freedman, & Phillips, 1990).Interest in strategies to reduce stress at work hasincreased steadily since the 1970s. According to theU.S. Department of Health and Human Services,national surveys conducted in 1985, 1992, and 1999found the prevalence of stress management and coun-seling programs among private-sector worksites inthose years was 27%, 37%, and 48%, respectively(Nigam, Murphy, & Swanson, 2003). The popularity
Katherine M. Richardson and Hannah R. Rothstein, De-partment of Management, Zicklin School of Business, Ba-ruch College, City University of New York.
Correspondence concerning this article should be addressedto Katherine M. Richardson, Baruch College, City Universityof New York, One Bernard Baruch Way, New York, NY10010. E-mail: [email protected]
Journal of Occupational Health Psychology2008, Vol. 13, No. 1, 6993
Copyright 2008 by the American Psychological Association1076-8998/08/$12.00 DOI: 10.1037/1076-8998.13.1.69
69
of worksite stress management programs has grownsignificantly abroad as well as in the United States.
Job Stress and Interventions
Newman and Beehr (1979, p.1) defined job stressas a situation wherein job-related factors interactwith the worker to change his or her psychologicaland/or physiological condition such that the person isforced to deviate from normal functioning. Implicitin this definition is the belief that work-related factorsare a cause of stress and that the individual outcomesmay be psychological, physiological, or some com-bination of these. A SMI may attempt to change thesework-related factors, assist employees in minimizingthe negative effects of these stressors, or both.Ivancevich et al. (1990) developed a conceptualframework for the design, implementation, and eval-uation of SMIs. According to the model, interven-tions can target three different points in the stresscycle: (a) the intensity of stressors in the workplace,(b) the employees appraisal of stressful situations, or(c) the employees ability to cope with the outcomes.The components of actual SMIs vary widely, encom-passing a broad array of treatments that may focus onthe individual, the organization, or some combination(DeFrank & Cooper, 1987; Giga, Noblet, Faragher,& Cooper, 2003).
Interventions may be classified as primary, sec-ondary, or tertiary. Primary interventions attempt toalter the sources of stress at work (Murphy & Sauter,2003). Examples of primary prevention programsinclude redesigning jobs to modify workplace stres-sors (cf. Bond & Bunce, 2000), increasing workersdecision-making authority (cf. Jackson, 1983), orproviding coworker support groups (cf. Carson et al.,1999; Cecil & Forman, 1990; Kolbell, 1995). Incontrast, secondary interventions attempt to reducethe severity of stress symptoms before they lead toserious health problems (Murphy & Sauter, 2003).Tertiary interventionssuch as employee assistanceprogramsare designed to treat the employeeshealth condition via free and confidential access toqualified mental health professionals (Arthur, 2000).The most common SMIs are secondary preventionprograms aimed at the individual and involve instruc-tion in techniques to manage and cope with stress(Giga, Cooper, & Faragher et al., 2003). Examplesare cognitivebehavioral skills training, meditation,relaxation, deep breathing, exercise, journaling, timemanagement, and goal setting.
Cognitivebehavioral interventions are designedto educate employees about the role of their thoughts
and emotions in managing stressful events and toprovide them with the skills to modify their thoughtsto facilitate adaptive coping (cf. Bond & Bunce,2000). These interventions are intended to changeindividuals appraisal of stressful situations and theirresponses to them. For example, employees aretaught to become aware of negative thoughts or irra-tional beliefs and to substitute positive or rationalideas (Bellarosa & Chen, 1997).
Meditation, relaxation, and deep-breathing inter-ventions are designed to enable employees to reduceadverse reactions to stresses by bringing about aphysical and/or mental state that is the physiologicalopposite of stress (cf. Benson, 1975). Typically, inmeditation interventions, the employee is taught tofocus on a single object or an idea and to keep allother thoughts from his or her mind, although someprograms teach employees to observe everything thatgoes through their mind without getting involvedwith or attached to them. Meditation interventionsoften also include relaxation therapy and deep breath-ing exercises. Relaxation therapy focuses on the con-scious and controlled release of muscle tension. Deepbreathing exercises focus on increasing the intake ofoxygen and the release of carbon dioxide, althoughmuscle and mental relaxation is often an additionalgoal of slowing and deepening the breath.
Exercise programs generally focus on providing aphysical release from the tension that builds up instressful situations, increasing endorphin production,or both, although some have the goal of focusing theemployees attention on physical activity (rather thanon the stressors) or providing an outlet for anger orhostility (cf. Bruning & Frew, 1987).
Journaling interventions require the employee tokeep a journal, log, or diary of the stressful events inhis or her life (cf. Alford, Malouff, & Osland, 2005).The journal is used as a means of assisting the em-ployee to monitor stress levels, to identify the recur-ring causes of stress, and to note his or her reactions.Journals are also used to formulate action plans formanaging stress.
Time management and goal-setting interventionsare designed to help people manage their time better,both on and off the job. Employees often operateunder time pressure and are required to work onmultiple tasks simultaneously. Working under suchconditions can be particularly stressful. Time man-agement interventions provide skills training in theareas of goal setting, scheduling and prioritizingtasks, self-monitoring, problem solving, delegating,negotiating, and conflict resolution (cf. Bruning &Frew, 1987; N. C. Higgins, 1986).
70 RICHARDSON AND ROTHSTEIN
Electromyogram (EMG) biofeedback training pro-vides participants with continuous feedback of mus-cle tension levels (cf. Murphy, 1984b). The objectiveis to consciously trigger the relaxation response andcontrol involuntary stress responses from the auto-nomic nervous system. Through the use of electronicfeedback measuring forehead muscle tension or handtemperature, patients receive feedback via audibletones or visual graphs on a computer screen. Duringthe process they learn to monitor their physiologicalarousal levels and physically relax their bodies, ac-tually changing their heart rate, brain waves, musclecontractions, and more.
SMIs may focus on any one particular strategyoutlined above, such as relaxation training (cf. Fiedler,Vivona-Vaughan, & Gochfeld, 1989; R. K. Peters,Benson, & Porter, 1977), or combine multiple com-ponents to create a comprehensive training regimenthat may include any of the following: cognitivebehavioral skills, meditation, assertiveness, socialsupport, and so forth (cf. Bruning & Frew, 1986,1987; de Jong & Emmelkamp, 2000; Zolnierczyk-Zreda, 2002). A trained instructor or counselor ineither small group or one-on-one sessions generallyteaches the methods, but techniques are sometimesself-taught with the aid of books and tapes (cf.Aderman & Tecklenberg, 1983; Murphy, 1984b;Vaughn, Cheatwood, Sirles, & Brown, 1989) andincreasingly via computers (cf. Hoke, 2003;Shimazu, Kawakami, Irimajiri, Sakamoto, & Amano,2005). Treatment programs are usually administeredover several weeks and may take place at the work-site or at an outside location. Depending on thetreatment method used, session length may vary from15-min breaks to all-day seminars.
Which Interventions Are Most Effective?
The effectiveness of SMIs is measured in a varietyof ways. Researchers may assess outcomes at theorganizational level (e.g., absenteeism or productiv-ity) or at the individual level, using psychological(e.g., stress, anxiety, or depression) or physiological(e.g., blood pressure or weight) measures. Given thewide array of stress management programs and out-come variables, there has been much debate in theliterature as to which interventions, if any, are mosteffective (Briner & Reynolds, 1999; Bunce &Stephenson, 2000; Caulfield et al., 2004; DeFrank &Cooper, 1987; Giga, Noblet, Faragher, & Cooper, 2003;Ivancevich et al., 1990; Mimura & Griffiths, 2002;Murphy, 1984a; Newman & Beehr, 1979; Nicholson,Duncan, Hawkins, Belcastro, & Gold, 1988; van der
Hek & Plomp, 1997). Some critics have claimed thatstudies in this area are inconclusive and based largelyon anecdotes, testimonials, and methodologically weakresearch (Briner & Reynolds, 1999; Ivancevich et al.,1990).
Newman and Beehr (1979) were among the firstresearchers to perform a comprehensive narrativereview of personal and organizational strategies forhandling job stress, examining both medical and psy-chological literature. They reported a significant lackof empirical research in the domain and challengedindustrial/organizational psychologists to bring theirexperience to the field of job stressemployee health(Newman & Beehr, 1979). During the 1980s, severalresearchers reviewed the SMI literature (DeFrank &Cooper, 1987; Murphy, 1984a; Nicholson et al.,1988). Murphy (1984a) performed a narrative reviewof 13 studies and concluded that they generally dem-onstrated acceptable positive effects, but noted thatthe reports varied significantly in terms of the ade-quacy of the methodology used (p. 2). DeFrank andCooper (1987) criticized the methodological qualityof the studies reviewed by Murphy. Three of the 13studies from that review did not use control groups,and the majority had small sample sizes with lowpower and short follow-up time frames, making itdifficult to assess the maintenance of gains.
Nicholson et al. (1988) expanded on Murphys(1984a) work and identified 62 published evaluationsof stress management programs from the medicine,public health, psychology, and education fields.These included case studies, preexperiments, quasi-experiments, and experimental studies (Nicholson etal., 1988). Of the 62 studies examined, only 18 pro-vided adequate data to quantitatively summarize theresults. Within these 18 studies, the average improve-ment in the treatment groups was equal to threequarters of one standard deviation in the controlgroup scores, yielding mildly encouraging results(Nicholson et al., 1988). This translates into a Cohensd of 0.75, which approaches a large effect (J. Cohen,1988). However, only a third of the studies in Nicholsonet al.s review included participants from occupationalsettings. The remainder consisted of participants from avariety of groups, including schoolchildren, juveniledelinquents, college students, epilepsy patients, hyper-tension patients, and substance abusers.
Over the past two decades, intervention studies inoccupational settings have proliferated, and research-ers have conducted more focused reviews to examinetheir effectiveness (Briner & Reynolds, 1999; Bunce& Stephenson, 2000; Caulfield et al., 2004; Giga, No-blet, Faragher, & Cooper, 2003; Mimura & Griffiths,
71EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
2002; van der Hek & Plomp, 1997). Results havecontinued to be mixed. Briner and Reynolds con-ducted a narrative review of organization-level inter-ventions (e.g., job redesign) and concluded that theyoften have little or no effect. Bunce andStephenson examined interventions that focused onindividual-level outcomes and reviewed 27 studies,assessing their levels of descriptive information andstatistical power. They reported that at present thequality of reporting and research design is such that itis difficult to form an impression of what type of SMIis appropriate to whom, and in what circumstances(p. 198).
More recent narrative reviews have focused onparticular job occupations, such as the nursing pro-fession (Mimura & Griffiths, 2002) or mental healthprofessionals (Edwards, Hannigan, Fothergill, &Burnard, 2002), or specific geographic regions,such as the United Kingdom (Giga, Noblet, Faragher,& Cooper, 2003) and Australia (Caulfield et al.,2004). A problem with such focused reviews, how-ever, is that they result in a small number of studies,and each study may compare multiple interventionmethods or outcomes. This makes it difficult to ob-tain precise estimates of the relative effectiveness ofdifferent interventions, or to assess generalizability ofresults.
The accumulation of stress management studiesacross a wide variety of occupational and geographicsettings, assessing a multitude of intervention tactics,calls for a systematic review. In behavioral and med-ical research, as well as in other fields, meta-analysishas become the widely accepted technique for assess-ing the effectiveness of interventions. It has replacedthe traditional narrative assessment of a body ofresearch as a better way to accumulate data andsynthesize them into generalizable knowledge (Eden,2002). The primary goals of meta-analysis are toderive the best estimate of the population effect sizeand to determine whether there are any sources ofvariance around this effect (Rothstein, McDaniel, &Borenstein, 2002).
A study by van der Klink, Blonk, Schene, and vanDijk (2001) used meta-analytic techniques to exam-ine the effectiveness of worksite stress interventions.The researchers meta-analyzed data from 48 inter-ventions published in 45 articles between 1977 and1996. A small but significant overall effect size wasfound (d! 0.34). When effects were broken down byintervention type, cognitivebehavioral interventionsachieved the largest effect size (d ! 0.68), followedby multimodal (d ! 0.51), relaxation (d ! 0.35), andorganization-focused programs (d ! 0.08). The
present study seeks to update the van der Klink et al.meta-analysis. We believe that a new meta-analysis isappropriate for three reasons. First, although thevan der Klink et al. review was published in 2001, itincludes studies published only through 1996, and aconsiderable number of new, methodologically rig-orous studies have been conducted since that time.There is a growing consensus that meta-analyses andsystematic reviews, particularly those with healthpolicy or practice implications, should be updatedwhenever a sizable number of new studies appear (cf.Chalmers & Haynes, 1994; Clark, Donovan, &Schoettker, 2006; J. P. T. Higgins & Green, 2005).This review expands the eligibility timeframethrough early 2006.
Second, van der Klink et al. (2001) included stud-ies of varying study designs and methodologicalquality. Nine of their included studies used quasi-experimental designs, and others did not include ano-treatment control or comparison group. In addi-tion, the van der Klink et al. meta-analysis includedseven randomized experiments that did not reportsufficient statistics to calculate an effect size or thatreported results of significant outcomes and not oth-ers. They did this by making assumptions about theprobability values in these studies. As one of theconsistent criticisms of studies of SMIs is that theyare methodologically weak, our goal was to focusonly on studies with methodologically strong de-signs. We therefore included only those interventionsthat were evaluated using a true experimental design,with random assignment of participants to treatmentand control groups. This reduces the threat of selec-tion bias and increases the internal validity of theincluded studies (Cook & Campbell, 1976) and,therefore, the validity of the meta-analytic results.
Finally, the van der Klink et al. (2001) review wasbased entirely on published journal articles and ex-cluded dissertations, conference proceedings, bookchapters, and the like. This exclusion of so-calledgray literature has been shown to increase thethreat that publication bias (a tendency toward prep-aration, submission, and publication of research find-ings that is based on the nature and direction of theresearch results rather than on the quality of theresearch; Dickersin, 2005) will affect the meta-analytic results. Publication bias affects the degree towhich published literature is representative of all theavailable scientific evidence; when the research thatis readily available differs in its results from theresults of all the research that has been done in anarea, we are in danger of drawing the wrong conclu-sion about that body of research (Rothstein, Sutton,
72 RICHARDSON AND ROTHSTEIN
& Borenstein, 2005). In order to minimize the threatof publication bias, the current meta-analysis in-cluded experimental studies from both published andunpublished sources.
Our meta-analysis had two main goals. First, it wasintended to bring together and synthesize experimen-tal studies of SMIs that have been conducted (andreported) in a variety of disciplines (e.g., education,health care, organizational studies, and psychology)to identify what works, how well it works, and whereor for whom it works. Second, we hoped to use ourmeta-analytic results to identify areas in the stressmanagement literature where additional primarystudies are needed.
To summarize, we intended our systematic reviewto improve on the van der Klink et al. (2001) meta-analysis by drawing from an additional decade ofstudies, by including only randomized experimentswith sufficient statistical data to compute effect sizeswithout making additional assumptions about proba-bility values, and by incorporating a more compre-hensive literature search strategy. Rather than ex-clude the seven randomized experiments from thevan der Klink et al. study for which they inferred pvalues, we performed a sensitivity analysis using theeffect sizes for these studies that were imputed byvan der Klink et al. to determine whether their inclu-sion changes our results. Our goals are to synthesizeresearch from a variety of disciplines to assess whathas been learned and also to identify areas in whichadditional primary studies are needed.
Method
Literature Search
Studies that assessed the effectiveness of a work-site SMI were collected from a variety of sources.Because our intent was to build on the van der Klinket al. (2001) meta-analysis, we began by obtaining all45 studies used in that review. Second, we conductedan electronic search of six databases: AcademicSearch Premier, British Library Direct, DissertationsAbstracts, ERIC, ProQuest ABI Inform Global, andPsycARTICLES. These databases were chosen toobtain studies from different countries in a broadrange of research fields, including social sciences,health care, and education. In addition, several ofthese databases include both published and unpub-lished studies. For the Academic Search Premier,British Library Direct, ERIC, and PsycARTICLESdatabases, we entered the following search terms onthree lines: employee or work or management, AND
stress or wellness, AND program* or intervention orprevention. This same procedure was followed forthe Dissertations Abstract database, but it yielded noresults. We therefore changed the search criteria totwo lines: worksite and stress and management, ANDprogram or intervention or prevention. For the Pro-Quest ABI Inform Global database, the same searchterms were used but were entered on one line withclassification codes, as follows: SU(employee orwork or management) AND ((stress or wellness) w/2(program* or intervention or prevention)) ANDCC(9130 or 5400). The classification codes 9130 and5400 indicate experimental and theoretical work,respectively. These electronic searches yielded942 documents (Academic Search Premier ! 377,British Library Direct ! 255, ERIC ! 122,PsycARTICLES ! 99, Dissertation Abstracts !31, and Proquest ABI Inform Global ! 58).
Third, we performed a network search and e-mailed colleagues knowledgeable in the field to see ifthey recommended any studies. We also attended theAmerican Psychological Association/National Insti-tute for Occupational Safety and Health Work,Stress, and Health 2006 conference and reviewed theconference proceedings. Fourth, we performed asnowball search and reviewed the reference list ofeach article obtained to identify additional citationsbeyond the electronic search. Finally, we searchedprivate and government-sponsored Web sites devotedto stress research to locate additional unpublishedliterature, including those for the American Institutefor Stress (www.stress.org), the Canadian Instituteof Stress (www.stresscanada.org), and the NationalInstitute for Occupational Health and Safety(www.cdc.gov/niosh).
Criteria for InclusionA study had to meet the following criteria to be
included in the meta-analysis: (a) be an experimentalevaluation of a primary or secondary SMI (i.e., em-ployee assistance programs were excluded); (b) in-clude participants from the working population (i.e.,studies involving students were excluded) who arenot already diagnosed as having a major psychiatricdisorder (e.g., depression or posttraumatic stress) orstress-related somatic disorder (e.g., hypertension);(c) use random assignment to treatment and controlconditions; (d) report sample sizes, means, and stan-dard deviations for both treatment and a no-treatmentor waiting-list control group; if means and standarddeviations were not reported, some other type ofstatistic that could be converted into a standardized
73EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
mean effect size (Cohens d) was necessary; and (e)be written in English after 1976. This date was cho-sen because it is the year the APA Task Force onHealth Research published a report that exhortedpsychologists, including industrial/organizationalpsychologists, to take a role in examining the healthproblems of Americans (Beehr & Newman, 1978).
Among the 45 articles used in the van der Klink etal. (2001) meta-analysis, 19 met the inclusion criteriaoutlined above and were included in our study. Theremaining 26 articles were excluded for the followingreasons: Nine used a quasi-experimental design with-out random assignment, 8 did not include a no-treatment comparison group, 7 did not report suffi-cient statistics to calculate an effect size, 1 used astudent sample, and 1 did not include outcome vari-ables related to stress or strain. We reviewed theabstracts of the 942 articles identified in the elec-tronic search online to determine whether to obtainthe studies for a full text review. Most studies wereexcluded on initial abstract review because theylacked a control group, did not include appropriateparticipants (e.g., used students or patients with clin-ically diagnosed disorders), assessed employee atti-tudes about interventions rather than the behavioraleffects of the interventions, or had already been iden-tified via the van der Klink et al. study. In total, 50experimental studies were obtained for a full textreview. Of these, 37 were excluded because theywere quasi-experimental (17), did not use a controlgroup (9), used a student or patient sample (7), or didnot include the appropriate statistics (4). Thirteenstudies met all the inclusion criteria and were in-cluded in the meta-analysis (9 peer-reviewed journalarticles and 4 unpublished dissertations). In addition,19 from the van der Klink et al. meta-analysis wereincluded. During this process, we identified 19 othernonempirical review articles and also obtained themto examine the reference lists for additional studies.
The network search resulted in four usable studies(one journal article, two book chapters, and one un-published dissertation). The snowball search resultedin 24 additional documents for review, and 2 of thesewere used in the study. The searches on the stresswebsites yielded no usable studies. In total, 38 arti-cles were included in the current meta-analysis, rep-resenting 36 separate studies and 55 interventions.
Coding Procedure
We coded the reports at five levels: study, treat-mentcontrol contrast, sample, outcome, and effectsize. Study-level coding recorded the articles full
citation, publication type, number of treatmentcontrol contrasts, and source of article (e.g., data-base). The treatmentcontrol contrast level was cre-ated because several of the studies evaluated morethan one type of treatment. Although some of thesewere totally independent evaluations, some had mul-tiple treatment groups (each with nonoverlappingparticipants) that were compared with a single con-trol group. We considered each of these treatmentcontrol contrasts as a separate comparison for codingpurposes. At this level of coding, we recorded infor-mation about the treatment and comparison groups,including program components (e.g., cognitivebehavioral skills training, meditation, exercise, etc.),where the treatment and comparison groups worked,where the program was delivered, who delivered thetreatment, and the duration of the program. We usedthis information to code the intervention type asprimary, secondary, or a combination. At the samplelevel, we coded the number of people in the sample(before attrition), the types of workers, and demo-graphic information including country of partici-pants. At the outcome level, we recorded each de-pendent variable and categorized them according topsychological measures (e.g., general mental health,anxiety, or depression), physiological measures (e.g.,diastolic and systolic blood pressure or pulse), andorganizational measures (e.g., productivity or absen-teeism). We also coded the type of measurementscale (e.g., continuous) and the source of the data(e.g., self-report). For each outcome variable, we thencoded the necessary information to calculate its effectsize, including treatment and comparison group sam-ple sizes, the statistical data, the direction of theeffect size, and whether this difference was reportedas statistically significant by the original investigator.We both coded the initial 10% of studies to assessintercoder reliability. Because there was close to100% agreement, Katherine M. Richardson coded theremaining studies, reviewing unclear data when nec-essary with Hannah R. Rothstein until a consensusdecision was reached.
Two of the studies that met our inclusion criteriaused groups as the unit of random assignment totreatment or control conditions, rather than assigningindividuals. This clustering of individuals withingroups reduces the effective sample size of thesestudies. As we have no basis for estimating thiseffect, or for adjusting the weights of these twostudies, we used the original sample sizes in ouranalyses. This should not have much of an impact onthe overall effect size estimates or on the moderatoranalyses, but it will inflate the statistical significance
74 RICHARDSON AND ROTHSTEIN
levels of these two studies and underestimate theconfidence intervals (CIs) for them.
Statistical Procedures
Comprehensive Meta Analysis Version 2 software(Borenstein, Hedges, Higgins, & Rothstein, 2005)was used to conduct the statistical analyses. Thestandardized mean difference (J. Cohen, 1992;Lipsey & Wilson, 2001) was calculated to representthe intervention effects reported in the eligible stud-ies. This effect size statistic is defined as the differ-ence between the treatment and control group meanson an outcome variable divided by their pooled stan-dard deviations. For this meta-analysis, the randomeffects model was most appropriate as heterogeneitywas expected owing to the variety of interventiontypes and occupational settings. Our procedures wereanalogous to a HunterSchmidt (1990) bare-bonesmeta-analysis in that we made no corrections forstatistical artifacts other than sampling error. Ourdecision was based on the lack of sufficient infor-mation in the retrieved studies to appropriatelymake such corrections (e.g., scale reliabilities). Bynot adjusting for artifacts, it is expected that thecalculated mean effect sizes will underestimatetheir actual values (Hunter & Schmidt, 1990). Wecalculated effect sizes at the treatment controlcontrast level. Our meta-analysis represents thecombined effect of 55 independent interventionswithin 36 studies. For studies that reported multi-ple outcome variables, all applicable effect sizesthat could be extracted were calculated and coded.However, we followed Lipsey and Wilsons (2001)advice that including multiple effect sizes from thesame intervention violates the assumption of inde-pendent data points that is fundamental to mostcommon forms of statistical analysis and inflatesthe sample size (N of effect sizes rather than N ofinterventions). They recommended either using theaverage of the outcomes at the treatment controlcontrast level to get a combined effect or selectingthe most representative outcome measure withineach module. Owing to the wide range of outcomevariables within our population of studies, we usedthe average of the outcomes at the interventionlevel for our overall analysis rather than introducesubjectivity into the analysis process by selectingwhat we felt would be most representative. Wealso looked at individual outcomes in a series ofsubgroup analyses. Each effect size was weightedby its precision, so that interventions with larger
samples contributed more to the estimate of thepopulation effect size.
Results
Demographics
Table 1 summarizes the studies selected for themeta-analysis. Thirty-eight articles were identifiedthat met the inclusion criteria, representing 36 sepa-rate studies and 55 interventions. Total sample sizewas 2,847 before attrition and 2,376 after attrition.Individual sample sizes after attrition ranged from 14to 219 participants, with a mean of 49 per interven-tion. The participants represented a wide range ofoccupations, including office workers, teachers,nurses and hospital staff, factory workers, mainte-nance personnel, and social services staff. Two thirdsof the studies were conducted in the United States,and the remainder represented a diverse range ofcountries, including Australia, Canada, China (Tai-wan and Hong Kong), Israel, Japan, the Netherlands,Poland, and the United Kingdom. Fifty-nine percentof the participants were female (based on 28 studies)and mean age was 35.4 (based on 18 studies). Aver-age intervention length was 7.4 weeks. The meannumber of treatment sessions was 7.5, each lasting anaverage of 12 hr.
Types of InterventionsThe studies primarily assessed secondary interven-
tion strategies to reduce the severity of an employeesstress symptoms. Only 8 studies included compo-nents that were considered primary intervention strat-egies, such as increasing workers decision-makingauthority (e.g., participatory action research) or socialsupport within the organization. All the interventionstudies compared at least one treatment group to ano-treatment or waiting list control. Fourteen studiesevaluated two treatment groups versus the same com-parison group, and 2 studies evaluated three treat-ment groups versus the same control. Each treat-ment control contrast was treated as a separateintervention for analysis purposes. The majority ofstudies (k! 24) evaluated interventions conducted ina group-training environment. Other modes of treat-ment included individual counseling sessions (k !3); self-taught techniques using the Internet, tapes, orbooks (k ! 5); or a combination of varying methods(k ! 4). Twenty-five studies (69%) included relax-ation and meditation techniques. Twenty studies(56%) included cognitivebehavioral skills training.
75EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
Tabl
e1
Prim
ary
Stud
ies
Incl
uded
inth
eM
eta-
Anal
ysis
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
entc
om
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
peA
derm
an&
Teck
lenbe
rg(19
83)
AV
ario
us
org
aniza
tions
(T!
21,C!
15)
Gene
rals
tress
educ
ation
sem
inar
,plu
smed
itatio
nan
dre
laxati
ontra
inin
gvi
aau
diot
apes
Anx
iety
(Ben
dig,1
956)
12w
eeks
Relax
ation
BV
ario
us
org
aniz
atio
ns
(T!
19,C!
15)
Gen
eral
stre
ssed
ucati
onse
min
arA
ltern
ative
Alfo
rd,M
alouf
f,&
Osla
nd(20
05)
ACh
ildpr
ote
ctiv
ese
rvice
soffi
cers,
Aus
tralia
(T!
31,C!
30)
Jour
nalw
ritin
gab
outr
ecen
tst
ress
reac
tions
and
emotio
ns
Gen
eral
men
talh
ealth
(GHQ
-12),
posit
ive
affe
ct,neg
ativ
eaf
fect,
jobsa
tisfa
ction
(JIG)
3da
ysA
ltern
ativ
e
Berto
ch,N
ielso
n,Cu
rley,
&Bo
rg(19
89)
ATe
ache
rs(T!
15,C!
15)
Hol
istic
prog
ram
:dee
pbr
eath
ing,
exer
cise,
relax
ation
,socia
lsupp
ort,
asse
rtive
ness
,and
nutri
tion
Stre
ss(D
SP),
stre
ss(O
SI),s
tress
(teach
erst
ress
mea
sure
),st
ress
(struc
tured
clini
cali
nter
view
)
12w
eeks
Mul
timod
al
Bond
&Bu
nce
(2000
)A
Offi
cew
ork
ers(
T!
24,
C!
20)
A
ccep
tance
com
mitm
ent
ther
apy
:cog
nitiv
e-be
havi
oral
skill
sto
enha
nce
emot
iona
lcop
ing
Gene
ralm
enta
lhea
lth(G
HQ-12
),dep
ress
ion
(Beck
),mot
ivati
on,
jobsa
tisfa
ction
,pr
open
sity
toin
nova
te
14w
eeks
Cogn
itive
-be
havi
oral
BO
ffice
work
ers
(T!
21,
C!
20)
In
nova
tion
prom
otio
npr
ogra
m:g
oals
ettin
g,pa
rticip
atory
actio
n,an
dpl
anni
ngto
enha
nce
prob
lem-fo
cuse
dco
ping
Org
aniza
tiona
l
Brun
ing
&Fr
ew(19
86,
1987
)A
Offi
cer
work
ers
(T!
16,
C!
16)
Man
agem
ents
kills
base
don
cogn
itive
-beh
avio
ral
tech
niqu
es,g
oals
ettin
g,tim
em
anag
emen
t,co
mm
unica
tion,
plan
ning
Pulse
,sys
tolic
&di
asto
licbl
ood
pres
sure
,galv
anics
kin
resp
onse
8w
eeks
Multi
moda
l
BO
ffice
work
ers
(T!
15,
C!
16)
Relax
ation
and
med
itatio
nte
chni
ques
Relax
ation
CO
ffice
work
ers
(T!
15,
C!
16)
Exer
cise
prog
ram
Alte
rnati
ve
(tabl
eco
ntin
ues)
76 RICHARDSON AND ROTHSTEIN
Tabl
e1
(conti
nued
)
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
entc
om
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
pe
Carso
net
al.(19
99)
AN
urse
s,U
nited
Kin
gdom
(T!
27,C!
26)
Socia
lsupp
ortg
roup
toen
hanc
eco
ping
abili
ties
Stre
ss(D
CL),
socia
lsu
ppor
t(SO
S),se
lf-es
teem
(RSE
S),em
otio
nale
xha
ustio
n(M
BI),
gene
ral
men
talh
ealth
(GHQ
)
5w
eeks
Org
aniz
atio
nal
Cecil
&Fo
rman
(1990
)A
Teac
hers
(T!
16,C!
19)
Stre
ssin
ocul
ation
train
ing
(Meic
henb
aum,
1977
)Sc
hool
stre
ss,t
ask-
base
dst
ress
,job
satis
facti
on,r
ole
over
load
,soci
alsu
ppor
t(all
SISS
),co
ping
skill
s
6w
eeks
Cogn
itive-
beha
vior
al
BTe
ache
rs(T!
17,C!
19)
Cow
orke
rsupp
ortg
roup
Org
aniza
tiona
l
Chen
(2006
)A
Offi
cew
ork
ers,
Isra
el(T
!24
units
,C!
13units
,N!
219
parti
cipan
ts)
Act
ive
learn
ing
expe
rienc
ede
signe
dto
incr
ease
parti
cipan
tspe
rson
alre
sourc
es
Socia
lsupp
ort(
House
,19
81),
perc
eived
contro
l(Ka
rasek
,19
79)
1w
eek
Alte
rnat
ive
Colli
ns(20
04)
AO
ffice
work
ers(
T!
9,C
!9)
Cogn
itive
-beh
avio
rals
kills
,co
mm
unica
tion,
relax
ation
tech
niqu
es,t
ime
man
agem
ent
Stre
ss(JS
I),bu
rnou
t(M
BI),
gene
ral
phys
icalh
ealth
,an
xiet
y(ST
AI)
5w
eeks
Multi
moda
l
BO
ffice
work
ers
(T!
8,C
!9)
Cogn
itive
-beh
avio
rals
kills
,re
laxati
onte
chni
ques
Mul
timod
al
deJo
ng&
Emm
elkam
p(20
00)
AV
ario
us
org
aniza
tions,
the
Net
herla
nds(
T!
45,C!
41)
Mus
clere
laxati
on,c
ogn
itive
-be
havi
oral
skill
s,pr
oblem
solv
ing,
asse
rtive
ness
train
ing
(taug
htby
clini
cal
psyc
holo
gist)
Anx
iety
(Dutc
hSTA
I),ge
nera
lmen
talh
ealth
(GHQ
),ge
nera
lph
ysica
lhea
lth,
socia
lsupp
ort(
SSI),
role
over
load
(OSQ
),job
diss
atisfa
ction
(OSQ
)
8w
eeks
Multi
moda
l
BV
ario
us
org
aniz
atio
ns,
the
Net
herla
nds(
T!
44,C!
41)
Mus
clere
laxati
on,c
ogn
itive
-be
havi
oral
skill
s,pr
oblem
solv
ing,
asse
rtive
ness
train
ing
(taug
htby
train
edpa
rapr
ofes
siona
ls)
Mul
timod
al
(tabl
eco
ntin
ues)
77EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
Tabl
e1
(contin
ued)
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
entc
om
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
pe
Fava
etal
.(19
91)
AA
rmy
offi
cers
(T!
20,C!
17)
Stre
ssTy
pe-A
beha
vior
redu
ction
prog
ram
,co
gniti
ve-b
ehav
iora
lsk
ills,
relax
ation
,self
-es
teem
enha
ncem
ent
Stre
ss(gl
obal
asse
ssm
ento
frec
ent
stre
ss),
stre
ss(PS
S),ge
nera
lmen
talh
ealth
(Kell
ners
Sym
ptom
Q),e
xer
cise
attit
ude
28w
eeks
Mul
timod
al
Fied
ler,V
ivon
a-V
augh
an,&
Goc
hfeld
(1989
)
AH
azar
dous
was
tew
ork
ers(
T!
31,C
!30
)
Prog
ress
ive
musc
lere
laxati
on,d
eep
brea
thin
g
Systo
licbl
ood
pres
sure
,di
asto
licbl
ood
pres
sure
,gen
eral
sever
ityin
dex
(SCL-
90)
9w
eeks
Rela
xat
ion
Gan
ster,
May
es,S
ime,
&Th
arp
(1982
)A
Offi
cew
ork
ers
(T!
36,C!
34)
Cogn
itive
-beh
avio
ral
skill
s(M
eiche
nbau
m,19
75),
prog
ress
ive
musc
lere
laxati
on
Anx
iety,
depr
essio
n,irr
itatio
n,ge
nera
lph
ysica
lhea
lth,
epin
ephr
ine,
nore
pine
phrin
e
8w
eeks
Multi
moda
l
Gild
ea(19
88)
AFo
sterc
are
agen
cyw
ork
ers(
T!
9,C
!8)
Stre
ss/an
ger
man
agem
entt
rain
ing:
cogn
itive
-beh
avio
ral
skill
s,jou
rnalin
g,re
laxati
on,e
duca
tion,
ange
rcontro
l
Systo
licbl
ood
pres
sure
,di
asto
licbl
ood
pres
sure
,dep
ress
ion,
gene
ralp
hysic
alhe
alth,
hosti
lity,
and
anxiet
y(al
lSCL
-90)
n/a
Mul
timod
al
BFo
ster
care
agen
cyw
ork
ers(
T!
8,C
!8)
Relax
ation
,journ
aling
Relax
ation
N.C
.Hig
gins
(1986
)A
Offi
cew
ork
ers(
T!
17,C!
18)
Relax
ation
,sys
tem
atic
dese
nsiti
zatio
nEm
otio
nale
xha
ustio
n(M
BI),
stra
in(PS
Q),
abse
ntee
ism
6w
eeks
Rela
xat
ion
BO
ffice
work
ers,
mix
ed(T!
18,C!
18)
Ratio
nale
mot
ive
ther
apy,
time
man
agem
ent,
goal
setti
ng,a
sser
tiven
ess
train
ing
Mul
timod
al
Hok
e(20
03)
AV
ario
usorg
aniza
tions
(T!
46,C!
53)
Twice
wee
kly
e-m
ails
that
teac
hde
epbr
eath
ing,
exer
cise,
hypn
osis,
journa
ling,
med
itatio
nan
dre
laxati
on,i
mag
ery
Stre
ss(PS
S),de
pres
sion,
anxiet
y,an
ger,
daily
hass
les
12w
eeks
Mul
timod
al
(tabl
eco
ntin
ues)
78 RICHARDSON AND ROTHSTEIN
Tabl
e1
(contin
ued)
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
ent
com
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
pe
Jack
son
(1983
)A
Hos
pital
work
ers(
N!
66)
Intro
ducti
onofs
taff
mee
tings
toin
crea
sest
affp
artic
ipati
on
Role
confli
ct,ro
leam
bigu
ity,s
ocia
lsu
ppor
t,ge
nera
lm
enta
lhea
lth(G
HQ),
jobsa
tisfa
ction
,ab
sent
eeism
24w
eeks
Org
aniza
tiona
l
Kol
bell
(1995
)A
Child
prot
ectiv
ese
rvice
swork
ers(
T!
13,C!
12)
Med
itatio
nan
dre
laxati
ontra
inin
gvia
audi
otap
es
Emot
iona
lexha
ustio
n(M
BI),
gene
ral
phys
icalh
ealth
(BSI)
,abs
entee
ism
4w
eeks
Rela
xat
ion
BCh
ildpr
ote
ctiv
ese
rvice
swork
ers(
T!
13,C!
12)
Socia
lsupp
ortg
roup
Org
aniza
tiona
l
Lee
&Cr
ocke
tt(19
94)
AH
ospi
talnurs
es,
Taiw
an,C
hina
(T!
29,C!
28)
Ass
ertiv
enes
stra
inin
gba
sed
on
ratio
nal-
emotiv
eth
erap
y(E
llis,1
962)
Ass
ertiv
enes
s(RA
S),st
ress
(PSS)
2w
eeks
Cogn
itive-
beha
vior
al
Mad
di,K
ahn,
&M
addi
(1998
)A
Offi
cew
ork
ers,
mid
-
level
man
ager
s(T
!18
,C!
16)
H
ardi
ness
train
ing,
ba
sed
on
cogn
itive
-be
havi
oral
tech
niqu
es
Har
dine
ss,jo
bsa
tisfa
ction
,stra
in,
gene
ralp
hysic
alhe
alth,
socia
lsupp
ort
10w
eeks
Cogn
itive
-be
havi
oral
BO
ffice
work
ers,
mid
-
level
man
ager
s(T
!12
,C!
16)
Relax
ation
and
med
itatio
nRe
laxati
on
Mur
phy
(1984
b)A
Hig
hway
main
tenan
cew
ork
ers(
T!
15,C
!8)
Elec
trom
yogr
aphi
cbi
ofee
dbac
kG
ener
alph
ysica
lhea
lthan
dan
xiet
y(B
SI);
trait
anxiet
y(ST
AI);
jobdi
ssati
sfacti
on.
10da
ysA
ltern
ative
BH
ighw
aym
ainte
nan
cew
ork
ers(
T!
11,C
!8)
Mus
clere
laxati
onvia
cass
ette
tape
sRe
laxati
on
Pete
rs,B
enso
n,&
Porte
r(19
77);
Pete
rs,
Bens
on,&
Pete
rs(19
77)
AO
ffice
work
ers,
Unite
dK
ingd
om(T!
54,
C!
36)
Dail
y15
-min
brea
ks,
taug
htsp
ecifi
cre
laxati
onte
chni
que
with
deep
brea
thin
g
Gen
eral
phys
icalh
ealth
(Symp
tomsI
ndex
),pr
oduc
tivity
,socia
lsu
ppor
t,ha
ppin
ess,
systo
lic&
dias
tolic
bloo
dpr
essu
re
8w
eeks
Rela
xat
ion
BO
ffice
work
ers,
Unite
dK
ingd
om(T!
36,
C!
36)
Dail
y15
-min
brea
ks,n
ore
laxati
onte
chni
ques
taug
ht
Relax
ation
(tabl
eco
ntin
ues)
79EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
Tabl
e1
(contin
ued)
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
entc
om
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
pePe
ters
&Ca
rlson
(1999
)A
Uni
versi
tym
ainten
ance
wor
kers
(T!
21,C
!19
)
Hea
lthed
ucati
on,
cogn
itive
-beh
avio
ral
skill
s,go
alse
tting
,an
dre
laxati
on
Anx
iety,
ange
r,an
dde
pres
sion
(STPI)
,he
alth
self-
effic
acy,
jobsa
tisfa
ction
,sy
stolic
/dias
tolic
bloo
dpr
essu
re,
chol
ester
ol
10w
eeks
Mul
timod
al
Prui
tt(19
92)
AA
rmy
pers
onne
l(T!
31,C!
33)
Stre
ssaw
aren
ess
educ
ation
,as
serti
vene
ss,t
ime
man
agem
ent,
relax
ation
via
audi
otap
es
Anx
iety
(STAI
),an
xiet
y(SC
L-90
),sy
stolic
&di
asto
licbl
ood
pres
sure
n/a
Mul
timod
al
Shap
iro,A
stin,
Bish
op,
&Co
rdova
(2005
)A
Hea
lthca
repr
ofes
siona
ls(T!
18,C!
10)
Min
dful
ness
-bas
edst
ress
redu
ction
(med
itatio
n,de
epbr
eath
ing,
yoga
)
Burn
out,
gene
ral
men
talh
ealth
(GSI)
,pe
rceiv
edst
ress
8w
eeks
Rela
xat
ion
Shar
p&
Form
an(19
85)
ATe
ache
rs(T!
30,C
!30
)St
ress
inoc
ulati
ontra
inin
g(M
eiche
nbau
m,19
77)
Anx
iety
(TQ4
),an
xiet
y(ST
AI,s
tate
),an
xiet
y(ST
AI,tr
ait)
4w
eeks
Cogn
itive-
beha
vior
al
BTe
ache
rs(T!
30,C
!30
)Cl
assro
omm
anag
emen
tsk
illst
rain
ing
Alte
rnati
ve
Shim
azu,
Kaw
akam
i,Iri
maji
ri,Sa
kam
oto,
&A
man
o(20
05)
AO
ffice
work
ers
ina
const
ruct
ion
mac
hine
ryco
mpa
ny,J
apan
(T!
100,
C!
104)
Self-
pace
donlin
ein
terve
ntio
nte
achi
ngco
gniti
ve-b
ehav
iora
lan
dco
ping
tech
niqu
es
Self-
effic
acy,
stre
ss(B
JSQ)
,gen
eral
phys
icalh
ealth
,job
satis
facti
on
11w
eeks
Cogn
itive
-be
havi
oral
Stan
ton
(1991
)A
Adm
inist
rativ
eoffi
cew
ork
ers,
Aus
tralia
(T!
15,C!
15)
Eg
o-en
hanc
emen
tre
laxati
onte
chni
ques
Stre
ss(s
tress
ther
mom
eter
)n/a
Relax
ation
Thom
ason
&Po
nd(19
95)
ACu
stodi
alst
aff(
T!
14,C!
13)
Cogn
itive
-beh
avio
ral
skill
s,re
laxati
on,
imag
ery,
and
self-
man
agem
ents
kills
Gen
eral
men
talh
ealth
(SCL-
90),
anxiet
y(ST
AI),
jobsa
tisfa
ction
(JIG)
,bl
ood
pres
sure
6w
eeks
Multi
moda
l
BCu
stodi
alst
aff(
T!
13,C!
13)
Cogn
itive
-beh
avio
ral
skill
s,re
laxati
on,a
nd
imag
ery
Mul
timod
al
CCu
stodi
alst
aff(
T!
14,C!
13)
Pers
onal
deve
lopm
ent
skill
sA
ltern
ative
(tabl
eco
ntin
ues)
80 RICHARDSON AND ROTHSTEIN
Tabl
e1
(conti
nued
)
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
entc
om
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
pe
Tsai
&Cr
ocke
tt(19
93)
AH
ospi
talnurs
es,
Taiw
an,C
hina
(T!
68,C!
69)
Relax
ation
train
ing
base
don
cogn
itive
-beh
avio
ral
mode
lofr
elaxa
tion
(Smith
,199
0)
Gen
eral
men
talh
ealth
(Chin
eseG
HQ)
,stre
ss(C
hinese
NSC
)
5w
eeks
Rela
xat
ion
Tunn
eclif
fe,L
each
,&Tu
nnec
liffe
(1986
)A
Teac
hers
,A
ust
ralia
(T!
7,C!
7)Co
llabo
rativ
ebe
havi
oral
consu
ltatio
nfo
cuse
don
prob
lem-so
lvin
gap
proa
ch
Teac
hers
tress
(TOS
Q)5
wee
ksCo
gniti
ve-
beha
vior
al
BTe
ache
rs,
Aust
ralia
(T!
7,C!
7)Re
laxati
ontra
inin
gRe
laxati
on
Vau
ghn,
Chea
twoo
d,Si
rles,
&Br
ow
n(19
89)
AA
dmin
istra
tive
work
ers(
T!
8,C
!10
)
Prog
ress
ive
musc
lere
laxati
onvia
audi
otap
esSt
ress
(SRI)
4w
eeks
Relax
ation
von
Baey
er&
Kra
use
(1983
-19
84)
AN
urs
esin
abu
rntre
atm
entu
nit,
Cana
da(T!
7,C
!7)
Cogn
itive
-beh
avio
rals
kills
,de
epbr
eath
ing,
relax
ation
,and
role
play
ing
Anx
iety
(STAI
,sta
it)an
xiet
y(ST
AI,tr
ait)
1w
eek
Multi
moda
l
Wirt
h(19
92)
ABa
kery
empl
oyee
s(T
!28
,C!
18)
Lead
er-fa
cilita
tedSM
I:co
gniti
ve-b
ehav
iora
lsk
ills,
emotio
nals
tress
man
agem
ent,
relax
ation
,he
alth
educ
ation
Locu
sofc
ontro
l,ge
nera
lph
ysica
lhea
lth(PS
C),
abse
ntee
ism
4w
eeks
Multi
moda
l
BBa
kery
empl
oye
es(T
!15
,C!
18)
Self-
taugh
tSM
I:co
gniti
ve-
beha
vior
alsk
ills,
relax
ation
,hea
lthed
ucati
on
Mul
timod
al
Yun
g,Fu
ng,C
han,
&La
u(20
04)
AA
dmin
istra
tive
nurs
ing
staf
f,H
ong
Kon
g,Ch
ina
(T!
17,C
!30
)
Relax
ation
usin
gst
retc
hing
and
relea
sing
ofm
usc
lesA
nxiet
y(C
hinese
STA
I,st
ate),
anxiet
y(C
hinese
STA
I,tra
it),ge
nera
lm
enta
lhea
lth(C
hinese
GH
Q)
4w
eeks
Rela
xat
ion
BA
dmin
istra
tive
nurs
ing
staf
f,H
ong
Kon
g,Ch
ina
(T!
18,C
!30
)
Relax
ation
usin
gco
gniti
veim
ager
yRe
laxati
on
(tabl
eco
ntin
ues)
81EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
Many of the interventions had multiple components(such as cognitive behavioral skills training andmeditation). Fourteen of the studies evaluated inter-ventions with four or more treatment components.Table 1 provides a summary of the included studies.
Outcome Variables
Each study contained multiple outcome mea-sures. We selected and coded all dependent vari-ables that related to stress, including psychologi-cal, physiological, and organizational outcomes.This resulted in more than 60 different outcomevariables, or an average of 3 4 outcomes perstudy. We averaged these outcomes at the treat-ment control contrast level to calculate a com-bined effect size for each intervention. Psycholog-ical measures were used in 35 out of 36 studies.The most common among these were stress (k !14), anxiety (k ! 13), general mental health (k !11), and job/work satisfaction (k ! 10). Unfortu-nately, there was no uniform scale used for anyconstruct. For example, stress was measured via 11different scales, including the Job Stress Index(Sandman, 1992), Occupational Stress Inventory(Osipow & Spokane, 1983), Perceived Stress Scale(Cohen, Kamarck, & Mermelstein, 1983), Personaland Organizational Quality Assessment (Barrios-Choplin & Atkinson, 2000), and Teacher StressMeasure (Pettegrew & Wolf, 1982).
Physiological measures were used in a quarter ofthe studies, and the most common of these was sys-tolic and diastolic blood pressure. Other physiologi-cal measures were epinephrine and norepinephrinelevels, galvanic skin response, and cholesterol. Onlysix studies measured organizational-specific out-comes. Four studies assessed absenteeism, and twoexamined productivity.
Effect SizesA combined analysis, using the inverse-variance
weighted average effect size from each individualintervention, yielded a significant effect size acrossall studies (d ! 0.526, 95% CI ! 0.364, 0.687). Thisis considered a medium to large effect size (J. Cohen,1988). By comparison, van der Klink et al.s (2001)meta-analysis yielded a small combined effect size(d ! 0.34, 95% CI ! 0.27, 0.41). We checked forheterogeneity of effects in two ways. First, we usedthe traditional chi-square statistic to test the hypoth-esis that all of the observed heterogeneity was due tosampling error variance. The Q value was highlyTa
ble
1(co
ntinu
ed)
Aut
hor(s
)and
year
Trea
tmen
t-con
trol
contr
ast
Sam
ple
sizea
and
desc
riptio
nbTr
eatm
entc
om
pone
nts
Out
com
esm
easu
redc
Leng
thIn
terv
entio
nty
pe
Zoln
ierc
zyk-
Zred
a(20
02)
AFi
nanc
ialse
ctor
offi
cew
ork
ers,
Polan
d(T
!40
,C!
45)
Cogn
itive
-beh
avio
rals
kills
,job
contro
l,so
cial
supp
ort,
asse
rtive
ness
,an
gerc
ontro
l
Prob
lem-fo
cuse
dco
ping
,em
otio
nal-f
ocus
edco
ping
,socia
lsupp
ort
10w
eeks
Mul
timod
al
Note
.T!
trea
tmen
t;C!
contr
ol;G
HQ-
12!
Gen
eral
Hea
lthQu
estion
naire
12;J
IG!
Job
inG
ener
alSc
ale;
DSP!
Der
ogat
isSt
ress
Profi
le;O
SI!
Occ
upat
iona
lStre
ssIn
vent
ory;
DCL!
DeV
illie
rsCa
rson
Lear
ySc
ale;
MB
I!M
asla
chB
urno
utIn
vent
ory;
SISS!
Stre
ssin
the
Scho
olSe
tting
;JSI!
Job
Stre
ssIn
dex;
STA
I!St
ate
Trai
tAnx
iety
Inve
ntor
y;SS
I!So
cial
Supp
ortI
ndic
ator
;OSQ!
Occ
upat
iona
lStre
ssQu
estion
naire
;PSS!
Perc
eive
dSt
ress
Scal
e;SC
L-90!
Sym
ptom
Chec
klist
90;P
SQ!
Pers
onal
Stra
inQu
estion
naire
;RA
S!
Rat
hus
Ass
ertiv
enes
sSc
ale;
STPI!
Stat
e-Tr
aitP
erso
nalit
yIn
vent
ory;
TQ4!
Teac
herQ
uesti
onna
ire4;
BJS
Q!
Brie
fJob
Stre
ssQu
estion
naire
;NSC!
Nur
seSt
ress
Chec
klist
;TO
SQ!
Teac
herO
ccup
atio
nalS
tress
Quest
ionna
ire;S
RI!
Stre
ssR
espo
nse
Inde
x;SM
I!st
ress
man
agem
enti
nter
vent
ion;
SOS!
Sign
ifica
ntO
ther
sSc
ale;
RSE
S!
Ros
enbe
rgSe
lf-Es
teem
Scal
e.a
Bas
edon
postt
reat
men
tmea
sure
s.b
U.S
.sam
ple,
unle
ssoth
erw
isenote
d.c
Freq
uent
lyuse
dsc
ales
note
din
pare
nthe
ses.
82 RICHARDSON AND ROTHSTEIN
significant (Q ! 202.6, p " .001), indicating thatthere was more heterogeneity of effects than could beaccounted for by sampling error. Second, we used theI2 statistic: I2! [(Q df)/Q]# 100%, where Q is thechi-square statistic and df is its degree of freedom(Higgins, Thompson, Deeks, & Altman, 2003). I2represents the amount of variability across studiesthat is attributable to between-study differencesrather than to sampling error variability. In this case,the I2 statistic suggests that 73% of the total varianceis due to between-study variance, or heterogeneity,rather than to sampling error. Both statistics sug-gested the likely presence of moderators, so we pro-ceeded to conduct subgroup analyses.
Intervention-Level Moderators
To search for moderators, we classified the inter-ventions into more homogeneous subgroups and per-formed analyses on these groupings, again using theaverage outcome effect size for each intervention,when more than one outcome was assessed. First, wecoded our interventions into the same categories usedin the van der Klink et al. (2001) meta-analysis:cognitivebehavioral, relaxation, organizational, ormultimodal (multiple component). Seven interven-tions could not be classified into these groupings.These studies evaluated interventions composed ofexercise or EMG feedback, journaling, personalskills development, or classroom management train-ing (for teachers). We therefore created an alterna-tive intervention category, while recognizing thatthis grouping did not represent a specific type ofintervention. Table 2 shows the average effect sizefor interventions in each of these categories.
The results in Table 2 show that the average effectsizes of both the cognitivebehavioral and the relax-ation intervention categories were larger than theanalogous average effects from the van der Klink et
al. (2001) meta-analysis. Somewhat surprisingly,there was a great deal of heterogeneity of effects inthe cognitivebehavioral intervention category (I2 !89.5, Q ! 57.0, p " .001). The alternative interven-tion category had the second largest average effectsize (d ! 0.909, 95% CI ! 0.318, 1.499), although italso had a wide confidence interval and, as would beexpected, substantial heterogeneity (I2 ! 85.3, Q !40.8, p " .001). Organizational interventions yieldedvirtually no effect, also consistent with the priormeta-analysis. Multimodal interventions, however,yielded a significant but small effect size in thepresent study (d ! .239, 95% CI ! 0.092, 0.386),which is lower than the van der Klink et al. study.The present meta-analysis included 15 studies (19interventions) with multimodal programs, whereasthe van der Klink et al. study included only 8 studiesin this category. Within such multimodal programs,cognitivebehavioral, relaxation, and even organiza-tional techniques can all be included as treatmentcomponents. Among the 19 multimodal interventionsin our meta-analysis, 5 included cognitivebehav-ioral components, 3 included relaxation components,and 11 included both cognitivebehavioral and re-laxation components. This makes it difficult to assesswhether the specific components, the mixture of com-ponents, the number of components, or a combina-tion of these factors is causing the intervention effect.Another way to categorize our studies is to look at thenumber of components per intervention. Table 3shows the effect sizes based on these subgroups.
The results in Table 3 suggest that interventionsthat focus on a single component are more effectivethan those that focus on multiple components. Thegeneral trend is that as each component is added, theeffect is reduced. However, there was significantheterogeneity among the one-component studies(I2 ! 81.6, Q ! 103.0, p " .001), and the results inTable 3 are confounded by intervention type. For
Table 2Cohens d and Confidence Intervals on the Basis of Intervention Type
Intervention type k N d 95% CI yCognitive-behavioral 7 448 1.164** 0.456, 1.871 .68*Relaxation 17 705 0.497*** 0.309, 0.685 .35*Organizational 5 221 0.144 $0.123, 0.411 .08Multimodal 19 862 0.239** 0.092, 0.386 .51*Alternative 7 455 0.909** 0.318, 1.499 N/ANote. k! number of interventions; d! combined effect size; CI! confidence interval; y!Cohens d from van der Klink et al. (2001).* p " .05. ** p " .01. *** p " .001.
83EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
example, the one-component interventions with thelargest effects were cognitivebehavioral interven-tions (k ! 2, d ! 1.230, 95% CI ! $0.968, 3.428).Four of the single-component interventions taughtsome form of personal development skills that spe-cifically related to increasing personal resources ormanagement skills that would assist employees intheir jobs (e.g., classroom management training).These yielded a significant large effect (d ! 1.154,95% CI! 0.332, 1.975). Seven used relaxation as thetreatment and obtained a significant medium effect(d ! 0.501, 95% CI ! 0.161, 0.841).
Among the studies with two treatment compo-nents, relaxation techniques were most likely to bethe primary treatment component. This was the casewith 10 interventions, yielding a significant mediumeffect size (d ! 0.502, 95% CI ! 0.265, 0.739).Seven interventions used cognitivebehavioral tech-niques as the primary treatment component, and theseyielded a significant large effect size (d ! 0.913,95% CI ! 0.320, 1.505). Among the studies withfour or more treatment components, cognitivebehavioral skills training and relaxation were likelyto be incorporated. Nine interventions included bothcognitive behavioral and relaxation components(d ! 0.296, 95% CI ! 0.086, 0.507), three includedcognitivebehavioral components (d ! 0.201, 95%CI ! $0.119, 0.522), and three included relaxationcomponents (d ! 0.215, 95% CI ! $0.069, 0.499).
We also examined whether treatment durationmade a difference among the interventions. The av-erage length was 7.4 weeks, but the range was 3 daysto 7 months. We therefore grouped studies accordingto the length of the intervention period (three studiesdid not provide this information). Because treatmentlength may be confounded with type of intervention,we further classified the studies based on interventiontype. Table 4 shows the effect sizes based on these
subgroups. The All studies column suggests thatshorter interventions are more effective than longerones. However, when one examines the data by in-tervention type, the pattern does not hold as firmly.The majority of interventions fall under the relax-ation and multimodal categories, but there are a lim-ited number of studies in each cell. Relaxation inter-ventions consistently produce medium-sized effects,regardless of length. Multimodal programs, in con-trast, appear to lose effect as treatment length in-creases.
Outcome-Level Moderators
Another way to classify the data into subgroups isto examine the outcome measures used in the studies.Outcome measures could be psychological, physio-logical, or organizational in nature, and we codedthem into 40 general categories. Some measures wereused more frequently than others. To assess whichmeasures produced larger effects and to examinewhether outcome variables differed between inter-vention types, we performed a subgroup analysis ofintervention type crossed with the outcomes we notedwere used most frequently. Table 5 shows the effectsizes based on these subgroups.
Disaggregating the data by outcome variable andtreatment type results in a small number of interven-tions in certain cells, but the analysis does illustrateseveral interesting findings. First, no studies thatevaluated single-mode cognitivebehavioral or orga-nizational interventions measured physiological out-comes. Thus, the large effect of the single-modecognitivebehavioral programs is based solely onpsychological and (less frequently) organizationalmeasures. Likewise, the large effects of alternativeinterventions are also based primarily on psycholog-ical variables. A pattern thus emerges in which out-
Table 3Cohens d and Confidence Intervals on the Basis of the Number ofTreatment Components
No. oftreatment
components k N d 95% CIOne 20 946 0.643*** 0.309, 0.977Two 18 970 0.607*** 0.346, 0.868Three 2 59 $0.104 $0.627, 0.418Four or more 15 716 0.271** 0.102, 0.440Note. k ! number of interventions; d ! combined effect size; CI ! confidence interval.** p " .01. *** p " .001.
84 RICHARDSON AND ROTHSTEIN
come variables are likely to be chosen on the basis ofthe type of intervention. For example, interventionsthat focus on the individual (e.g., cognitivebehavioral, journaling, and stress education) use psy-chological measures, and those that focus on organi-zational changes generally include organizationalmeasures. The result is that we are left with gaps inthe body of research. We are unable to assess whetheralternative treatment programs (e.g., journaling, ex-ercise, and personal coping skills)which have alarge effect on psychological and physiological out-comesproduce similar results using organizationalmeasures. Regarding specific organizational out-comes, measures of productivity appear to producelarger effects than absenteeism, but there is a generallack of studies that use such measures. In general,
there are no uniform outcome measures used to as-sess the effectiveness of stress management pro-grams. Only one third of the interventions reportedusing an actual measure of stress, and among thesethere was significant variation in the scales chosen.
To obtain a more pure assessment of whether typeof outcome measure played a moderating role, weselected only those studies for which psychologicaloutcomes were provided in combination with eitherphysiological or organizational measures. In otherwords, the same samples of participants were beingmeasured using both types of outcomes. All of thepsychological measures were based on self-report,continuous scales. This may affect the reliability ofthe outcomes. The physiological measures, in con-trast, were more likely to be administered by an
Table 4Cohens d on the Basis of Length of Treatment and Intervention Type
Length oftreatment All studies
Treatment type
CB Relax Org Multi Alternative
14 weeks 0.804*** (15) 1.477** (2) 0.560* (5) $0.097 (1) 0.274 (3) 1.217* (4)58 weeks 0.396*** (22) 1.576 (2) 0.500*** (7) $0.003 (2) 0.291* (9) 0.585* (2)912 weeks 0.346* (10) 1.230 (2) 0.315 (3) N/A 0.089 (4) 0.250 (1)%12 weeks 0.401** (4) 0.323 (1) N/A 0.328 (2) 0.718* (1) N/ANote. Numbers in parentheses represent total number of interventions. CB ! cognitive-behavioral; Relax ! relaxation;Org ! organizational; Multi ! multimodal.* p " .05. ** p " .01. *** p " .001.
Table 5Cohens d on the Basis of Outcome Variables and Intervention Type
Outcome variable All studies
Treatment type
CB Relax Org Multi Alternative
PsychologicalAll combined 0.535*** (52) 1.154** (7) 0.507*** (16) 0.134 (5) 0.258** (18) 0.905** (6)Stress 0.727*** (18) 1.007** (5) 0.834** (5) $0.314 (2) 0.595** (5) 1.367*** (1)Anxiety 0.678*** (22) 2.390*** (1) 0.611*** (5) N/A 0.418** (12) 0.841 (4)Mental health 0.441*** (16) 0.708* (1) 0.405* (5) 0.167 (3) 0.518 (5) 0.616* (2)Work-related outcomesa 0.183 (23) 0.682 (4) $0.381 (4) 0.243 (4) $0.115 (7) 0.794 (4)
PhysiologicalAll combined 0.292* (14) N/A 0.312 (5) N/A 0.166 (7) 0.714** (2)
OrganizationalAll combined 0.267 (11) 0.606 (1) 0.534*** (4) 0.247 (3) $0.122 (3) N/AProductivity 0.703*** (4) 0.606 (1) 0.661*** (2) 0.989** (1) N/A N/AAbsenteeism $0.059 (7) N/A 0.213 (2) $0.159 (2) $0.122 (3) N/A
Note. Numbers in parentheses represent total number of interventions. CB ! cognitive-behavioral; Relax ! relaxation;Org ! organizational; Multi ! multimodal.a Includes job/work satisfaction, motivation, social support, daily hassles, role ambiguity, role overload, and perceivedcontrol.* p " .05. ** p " .01. *** p " .001.
85EFFECTS OF STRESS MANAGEMENT INTERVENTIONS
independent, trained examiner or via a medical de-vice (e.g., blood pressure monitor). The organiza-tional measures were based on company records(e.g., absenteeism) and self-report data (e.g., produc-tivity or propensity to innovate). Table 6 shows theeffect sizes based on these subgroups.
The data in Table 6 are based on a small number ofstudies and should therefore be interpreted cau-tiously. Even so, it appears that intervention typecontinues to confound outcome effects. The resultsfor all studies combined suggest that psychologicaland physiological outcome variables produce compa-rable effect sizes. However, this depends on type ofintervention. For example, for the alternative sub-group, the physiological measures produced largereffect sizes. When studies measured both psycholog-ical and organizational variables, psychological out-comes produced larger effects than organizationaloutcomes. But this may also depend on treatmenttype, as the relationship is reversed for the relaxationand organizational subgroups. Table 6 does alert us tothe overreliance on self-report measures in interven-tion studies. Among the 55 interventions, only 11measured both psychological and physiological out-comes, and 11 measured both psychological and or-ganizational outcomes.
Sample-Level Moderators
The final moderator analysis we performed wasbased on industry sector. Many early interventionstudies were performed in the health care or educa-tion fields. We classified studies into subgroups onthe basis of three industry sectors: office, health care,and education. We further grouped the results by
intervention type. Table 7 shows the effect sizesbased on these subgroups. Results depict effect sizesin the general direction of prior analyses, with cog-nitive behavioral producing the largest effects.However, what may be more interesting is to exam-ine the distribution of intervention type by industry.For example, multimodal interventions appear morelikely to be used in office settings, perhaps becausethere has been no solid empirical evidence as to themost effective treatment in this particular setting, andtherefore a potpourri approach is used. In contrast,relaxation interventions appear more often withinhealth care settings, and cognitivebehavioral inter-ventions in education settings. However, disaggregat-ing the data produces small numbers of studies ineach cell, so these interpretations may be unstable.
Outlier Analysis
We performed outlier analyses by examining for-est plots of the effect sizes and confidence intervals,for all studies combined and at the subgroup levels.One alternative intervention was identified as a pos-sible outlier because of its very large effect size andthe fact that its confidence interval did not fall intothe range of similar interventions for that subgroup.We therefore excluded this study (which representedtwo alternative interventions: exercise only and lis-tening to music; Taylor, 1991) from the analysis.However, based on a sensitivity analysis, we notethat if we included this study in our calculations, thecombined overall effect size would increase slightly(d ! 0.618, 95% CI ! 0.432, 0.903).
Table 6Cohens d on the Basis of Outcome Variable and Intervention Type for Selected Studies
Outcome variable All studies
Treatment type
CB Relax Org Multi Alternative
Psychological vs.physiological
Psychological 0.227* (11) N/A 0.303* (4) N/A 0.151 (6) 0.250 (1)Physiological 0.219 (11) N/A 0.282 (4) N/A 0.115 (6) 0.601 (1)Psychological vs.
organizationalPsychological 0.285** (11) 0.253 (1) 0.376* (4) 0.187 (3) 0.197 (3) N/AOrganizational 0.267 (11) 0.606 (1) 0.534*** (4) 0.247 (3) $0.122 (3) N/ANote. Numbers in parentheses represent total number of interventions. CB ! cognitive-behavioral; Relax ! relaxation;Org ! organizational; Multi ! multimodal.* p " .05. ** p " .01. *** p " .001.
86 RICHARDSON AND ROTHSTEIN
Sensitivity Analysis
We performed a sensitivity analysis to examinewhether including seven additional studies from thevan der Klink et al. (2001) meta-analysis wouldchange our overall results. These particular studiesdid not meet our inclusion criteria because, accordingto our review, they did not report sufficient statisticsto calculate an effect size.1 They were included in thevan der Klink et al. study on the basis of assumptionsmade by those authors. However, rather than excludethese interventions entirely, we used the reportedeffect sizes from the van der Klink et al. study andadded them to our meta-analysis. This slightly re-duced our combined overall effect size (d ! 0.469,95% CI ! 0.328, 0.609) but still yielded a mediumeffect.
Publication Bias
We performed an analysis to examine whether thecombined effect size from published studies differedfrom that of unpublished studies. The current meta-analysis included five unpublished dissertations, rep-resenting eight interventions. The combined effectsize from these eight interventions, using the averageeffect size across outcomes, yielded a significantmedium to large effect (d ! 0.553, 95% CI !$0.126, 1.232, p ! .110). In comparison, the com-bined effect from published studies was .509 (95%CI ! 0.356, 0.661, p " .001). In this case, it appearsthat including unpublished studies slightly increasedthe size of our overall average effect, whereas thegeneral concern is that omission of unpublished workupwardly biases the effect size. We have no unam-biguous explanation for the direction of difference inour particular case. There were no apparent differ-ences in methodological quality between the twogroups of studies.
Another way to detect publication bias in a meta-analysis is the trim-and-fill technique, developedby Duval and Tweedie (2000). Trim and fill is anonparametric method designed to estimate and ad-just a funnel plot for the number and outcomes ofmissing studies (Duval, 2005). We used this methodon the overall effect size distributions and estimatedthe number of missing studies at six, all to the rightof the mean. This suggests that the combined effectof 0.526 is understated and that the potential impactof including the proposed missing studies wouldincrease the effect to 0.595. However, one limitationof the trim-and-fill method is that if the data areheterogeneous, the technique may impute studies thatare not really missing. Study-related factors maydistort the appearance of the funnel plot (Duval,2005). To adjust for heterogeneity, we performedadditional trim-and-fill analyses at the subgrouplevel, examining intervention type crossed with out-come. We were able to use the method only if therewas a minimum of three interventions in the sub-group. On the basis of the analyses, multimodal in-terventions was the only category that producedgreater than one missing study. On the basis of thepsychological outcome measures, it appeared thatthree studies were missing on the left side of themean, which would suggest an overstated effect andwould decrease the overall effect for multimodalinterventions (d ! 0.190). We stress that the maingoal of the trim-and-fill method is as a sensitivity
1 For example, one study contained three distinct treat-ment modules but combined the participants of each intoone group and compared it with the control. Two studiesreported results of only the statistically significant findings,and several others failed to report all required statistics (e.g.,means with no standard deviations). In such cases, van derKlink et al. (2001) used p values, making assumptions whenno such value was reported (e.g., nonsignificant findings).
Table 7Cohens d Based on Industry Sector and Intervention Type
Industry All studies
Treatment type
CB Relax Org Multi Alternative
Office 0.680*** (19) 0.872 (3) 0.619*** (7) 0.162 (1) 0.395* (6) 1.337** (2)Health care 0.492*** (8) 0.988*** (1) 0.462*** (4) 0.208 (2) 1.307* (1) N/AEducation 1.255** (7) 1.662* (3) 1.523* (1) 0.056 (1) 0.524 (1) 2.037*** (1)Note. Numbers in parentheses represent total number of interventions. CB ! cognitive-behavioral; Relax ! relaxation;Org ! organizational; Multi !