In 2006, more than 6 million violent crimes were committed in the United States,Reference Klaus and Matson1 and nearly 90% of people are exposed to at least 1 traumatic type of event during their lifetime.Reference Breslau2 Moreover, current estimates suggest that 5% to 10% of those exposed to traumatic events will meet diagnostic criteria for posttraumatic stress disorder (PTSD).Reference Ozer, Best, Lipsey and Weiss3 These daunting figures attest to the regrettable exposure and potential suffering that traumatic events may cause. Recognition is growing, however, that most people exposed to traumatic or stressful life events do not experience serious disruptions in normal life functioning.Reference Mancini and Bonanno4 These individuals may be considered to be exhibiting what is commonly referred to as resilience. Resilience may be thought of as the ability to adapt to or rebound from adversity. Therefore, it seems that there would be value in identifying the operational or tactical elements of human resilience so that interventions or training programs designed to enhance resilience may be better guided or informed. This article represents an attempt to test a foundational hypothesis regarding human resilience. More specifically, we test the hypothesis that the effects of stressor conditions upon select measurable outcomes such as burnout and job-related variables may be mediated through the cognitive and affective registrations (referred to as stress arousal herein) of those events in addition to any direct effects those conditions may exert.
RESILIENCE
Reivich and Shatte define resilience as the ability to “persevere and adapt when things go awry,” and suggest that the construct is associated with one's ability to cognitively appraise a situation.Reference Revich and Shatte5 Bonanno defines resilience as “the ability of adults in otherwise normal circumstances who are exposed to an isolated and potentially disruptive event, such as the death of a close relation or a violent or life-threatening situation, to maintain relatively stable, healthy levels of psychological and physical functioning.”Reference Bonanno6 Kaminsky et al propose a tripartite model of disaster mental health that breaks down resilience into resistance, which they conceptualize broadly as protective factors or immunity, and resilience that they define as “the ability of an individual, a group, an organization, or even an entire population, to rapidly and effectively rebound from psychological and/or behavioral perturbations associated with critical incidents, terrorism, and even mass disaster.”Reference Kaminsky, McCabe, Langlieb and Everly7
Empirical Evidence of Stressful Life Events, Disasters, and Resilience
Researchers have investigated various protective factors that foster resilience in people exposed to stressful life occurrences, with the bulk of this formative research being conducted on grieving individuals following the loss of a loved one.Reference Shuchter and Zisook8Reference Stein, Folkman, Trabasso and Richards9Reference Wortman and Silver10 In a longitudinal study by Bonnano and colleagues that began several years before the death of a spouse and continued for several years after the death of the spouse, 46% of the participants demonstrated no clinical depression at any time during the study.Reference Bonanno, Wortman and Lehman11 Moreover, in a study that compared younger (younger than 65 years) bereaved adults after the death of a spouse or a child with a matched group of nonbereaved individuals (those with intact marriages), 52% of the bereaved individuals were considered resilient (defined as scoring within 1 standard deviation of the nonbereaved group's mean on assessed symptom levels) at both 4 and 18 months postloss.Reference Bonanno, Moskowitz, Papa and Folkman12 In fact, by 18 months postloss, resilient individuals were equivalent to nonbereaved individuals on the symptom ratings.Reference Bonanno, Moskowitz, Papa and Folkman12
Some researchers have explored the resilience of individuals exposed to disasters, such as the aftermath of the terrorist attacks of September 11, 2001.Reference Bonanno, Galea, Bucciarelli and Vlahov13Reference Fraley, Fazzari, Bonanno and Dekel14Reference Fredrickson, Tugade, Waugh and Larkin15 In a study that used a representative sample of 2752 New York residents, more than 65% of the participants met the criteria for resilience, defined as having no or only 1 PTSD symptom.Reference Bonanno, Galea, Bucciarelli and Vlahov13 Moreover, more than half of the sample involved in the rescue efforts at the World Trade Center was resilient, as was close to 54% of the sample who experienced the death of a friend or relative as a result of the attacks.Reference Bonanno, Galea, Bucciarelli and Vlahov13 Another study reported that highly secure individuals tended to be better adjusted than dismissive individuals after the tragedy, as assessed by self-reported PTSD symptoms.Reference Fraley, Fazzari, Bonanno and Dekel14 The authors further suggested that secure or attached individuals actually may be able to experience some type of personal growth or internal strength after the attacks. Moreover, in a study of 46 college students, positive emotions served to buffer against depression after the attacks.Reference Fredrickson, Tugade, Waugh and Larkin15
Factors That Influence Resilience
Several factors have been purported to influence the enhancement of resilience. In a phenomenological, qualitative study of adult female survivors of childhood sexual abuse, resilience determinants (specific innate and learned characteristics that contribute to participants' ability to become resilient adults) and resilience processes (how participants in the study described becoming resilient) were assessed. Their results identified 5 resilience determinants: being interpersonally skilled or having the ability to interact positively and effectively with others; being competent (eg, excelling in school or athletics); having high interpersonal self-worth or self-regard; being spiritual; and having what may be perceived as helpful life circumstances (eg, being the youngest in the family). Resilience processes were reportedly enhanced by coping strategies (eg, writing, praying, keeping busy, setting boundaries), refocusing and moving on, healing actively (ie, taking responsibility for one's own recovery and refuting the “victim” role, often through counseling), and being able to integrate the trauma into their current life stories without excessive emotional discomfort, or active closure.Reference Bogar and Hulse-Killacky16
Factors That Influence Resilience, Job Satisfaction, and Job Performance
Studies have begun to explore the construct of resilience on emotional responsiveness and its impact on job performance. In 1 study, the relation between human resources, also referred to as psychological capital (broadly theorized as a positive state of development and includes dimensions including hope, optimism, and resilience), and job performance was assessed in 3 factories located in the People's Republic of China.Reference Luthans, Avolio, Walumbwa and Li17 Results indicate that psychological capital in the aggregate and hope, optimism, and resilience when considered separately correlated positively with performance (eg, a factory with performance outcome and relative merit-based salary).Reference Luthans, Avolio, Walumbwa and Li17 A later study confirmed these findings in that employees' hope, optimism, and resilience, when assessed both separately and combined in the construct of psychological capital, had a positive impact on job performance, organizational commitment, and organizational citizenship behavior.Reference Zhong18
The relation between positive emotionality, conceptualized as the presence of ambient positive emotions and the ability to express positive emotions during adversity, and measures of burnout, job satisfaction, perceived performance, and intention to leave one's job in a sample of certified public accountants were analyzed. The results of the study supported a complementary relation between positive emotions and negative emotions such that positive emotions appear to support job satisfaction and performance, whereas negative emotions appear to predict burnout and intentions of leaving the job.Reference Everly, Smith and Welzant19
The purpose of the present study was to expand upon these findings. In particular, the goal was to determine whether assessing and intervening at the cognitive-affective domain (stress arousal) would best foster resilience regarding burnout, job performance, and job outcome. More specifically, this study assessed whether burnout and job-related outcomes are predicted more by stress arousal (a cognitive-affective reaction) than by conditions associated with the job. Based on the recommendations of RodgersReference Rodgers20 and practices by Smith et alReference Smith, Davy and Everly21 and Everly et al,Reference Everly, Smith and Welzant19 structural modeling is used to provide greater support for arguments supporting causality than have past correlation analyses.
METHODS
Subjects
Subjects were selected from a database of responses from 701 individuals who were sampled from a mailing list provided by the American Institute of Certified Public Accountants. The database contained responses to a demographic data sheet and a battery of psychometric instruments. This study incorporated the responses of the 491 individuals in the aforementioned sample used in public accounting. Of these subjects, 58% (283) were men, 79% (387) were married, 95% (457) were white, and 61% (299) indicated that they were between 26 and 45 years old. More than 64% (314) possessed a bachelor's degree and another 34% (166) held a masters degree.
Measures
The following 3 role stress measures were incorporated into this study:
1. Role ambiguity: 3 items from the 14-item Role Conflict and Role Ambiguity ScaleReference Rizzo, House and Lirtzman22
2. Role conflict: 3 items from the 14-item Role Conflict and Role Ambiguity ScaleReference Rizzo, House and Lirtzman22
3. Role overload: 5 items from the Beehr et al scaleReference Beehr, Walsh and Taber23
Each of these measures has been identified as a significant workplace stressor.Reference Smith, Davy and Everly21 Fogarty et al discuss the acceptability of these measures in terms of their psychometric properties reported in prior research.Reference Fogarty, Singh, Rhoads and Moore24 Each of these constructs is measured on a 5-point Likert-type scale. The key outcome measures were as follows:
1. Performance: a 6-item scale drawn from Dubinski and MattsonReference Dubinski and Mattson25
2. Turnover intentions: 3 items drawn from Donnelly and IvancevichReference Donnelly and Ivancevich26
3. Job satisfaction: 27 items drawn from the Churchill et al scaleReference Churchill, Ford, Hartley and Walker27
4. Burnout: 24 items drawn from the multidimensional role-specific version of the Maslach Burnout InventoryReference Singh, Goolsby and Rhoads28
Each of the outcome measures were measured using 5-point Likert-type scales.
Stress arousal was measured using 17 items from the stress arousal scale (SAS).Reference Everly, Sherman and Smith29 This instrument was designed to tap the respondent's cognitive-affective domain (precipitators of the physiological stress response), which allows an indirect measure of one's level of stress arousal. The conditions that define emotional arousal (as measured by SAS) have been shown to be highly correlated with stress-related physical symptoms.Reference Lazarus and Folkman30Reference Everly and Sobelman31
The SAS has been used in a number of accounting research studies.Reference Smith, Davy and Everly21Reference Smith, Davy and Stewart32 Factor analysis on a large data set indicated that there were 2 underlying dimensions: psychological discord (13 items, α = .91) and relaxation (4 items, α = .86).Reference Smith, Everly and Johns33 Psychological discord was defined as “the state of emotional distress experienced as a result of cognitive interpretation of environmental events” and relaxation was defined as “a state of cognitive-affective psycho-physiological homeostasis, ie, the lack of extraordinary arousal.” These results have been replicated numerous times.Reference Smith, Davy and Everly21Reference Smith, Davy and Stewart32 Responses were made on 4-point Likert-type scales ranging from “seldom or never” (1) to “almost always” (4).
Analysis
A series of EQS structural modeling analyses (Multivariate Software, Encino, CA) were used to test the hypothesis that a putative causal model that included stress arousal, as measured by cognitive-affective indicia, would explain greater variance than a direct stressor to outcome effects model. The selection of this methodology was guided by the recommendation of Rodgers,Reference Rodgers20 who cogently argued that traditional null hypothesis significance testing yields less valuable information than does structural modeling to researchers in the behavioral sciences. He argued that structural modeling has the advantages of forcing theoretical precision, promoting more useful data analyses, and being more readily translated into practical applications.
Although the stress arousal, burnout, and job satisfaction scales were multifactorial in nature, the role stressor, performance, and turnover intentions scales were unidimensional (ie, the items on each scale loaded on a single factor). To facilitate the ensuing measurement model tests, we combined the items for each of these scales onto 2 composite indicator variables using a procedure described by Bentler and Wu.Reference Bentler and Wu34 This procedure is suitable when there is no expectation that any of the composites created would be different from each another, and “each composite should measure the same construct, or combination of constructs, as measured by a single composite of all the original scores.”Reference Bentler and Wu34 This procedure facilitated the development of a latent variable model by allowing for a better estimate of the random error associated with these constructs. Random error is taken into account when estimating paths from constructs to indicator variables and within the structural model.
We conducted a confirmatory factor analysis on the sample data to independently test the construct and discriminant validity among the constructs represented by the measures. Anderson and Gerbing prescribe assessment of the measurement model before testing the structural linkages.Reference Anderson and Gerbing35 The complete measurement model was tested using the elliptical estimation procedure in EQS version 6.1. Table 1 presents the items that comprised each latent variable to be tested along with the mean score and standard deviation for each predicted latent variable.
TABLE 1 Factors for Measurement Model Tests
We then conducted a series of EQS structural modeling analyses to assess the impact of stress arousal. The first model examined only the direct effects from the role stressors to the outcome constructs. The second model inserted stress arousal as a mediator in the relations between the role stressors and the outcomes. Then, in each analysis, we dropped statistically nonsignificant parameters based on the output of Wald tests applied to each model.Reference Bentler36 We assessed model fit for both the measurement and structural model tests using a variety of fit measures outlined by Bentler.Reference Bentler37 These measures include the goodness-of-fit chi square, the normed fit index, the non-normed fit index, the comparative fit index, the LISREL (Scientific Software International, Lincolnwood,IL) goodness-of-fit index, the average off-diagonal squared residual (AOSR), and the root mean square error of approximation (RMSEA). We assessed model fit using multiple measures because no single measure is definitive.Reference Fogarty, Singh, Rhoads and Moore24
RESULTS
Tables 2 and 3 present the measurement model test results. Table 2 indicates that the path coefficients from each latent construct to its manifest indicator is significant at P < .01. The fit indices reported in Table 3 indicate good model fit because each is above the 0.900 minimum threshold. In addition, the AOSR of 0.031 and the RMSEA of 0.07 fall within their standard of acceptance.
TABLE 2 Standardized Measurement Coefficients for the Construct Indicators
TABLE 3 Results of Measurement Model Tests—Panel B: Goodness-of-Fit Summary
Table 4 and Figure 1 present the results from testing the model, which examined the direct effects of the role stressors on the 4 key outcomes with stress arousal excluded from the analysis. The fit indices reported in Table 4 indicate a good model fit because each of their values is above the minimum threshold of 0.900. In addition, the AOSR value of 0.043 and the RMSEA of 0.075 fall within their respective standards for acceptance. As Figure 1 illustrates, role ambiguity has a significant positive influence on burnout (0.628) and turnover intentions (0.561) and has a significant negative influence on job satisfaction (−0.824) and performance (−0.251). Thus, role ambiguity appears to increase burnout and turnover intentions while decreasing job satisfaction and performance. Role overload has a significant positive influence on burnout (0.340) and turnover intentions (0.144). It does not affect job satisfaction or performance. Finally, role conflict did not have significant effects on any of the outcome constructs.
FIGURE 1 Direct Effects Model Standardized Path Coefficients
Paths between each latent construct and its indicators are omitted for ease of diagramming and interpretability. *Covariance between independent factors (all significant at P < .01). **Significant at P < .01.
TABLE 4 Direct Effects Model Goodnessof-Fit Test Results
Table 5 and Figure 2 present the results from testing the model, which includes stress arousal as a posited mediator in the relations between sources of role stress and the 4 key outcomes. With the exception of the LISREL goodness-of-fit index, the fit indices reported in Table 5 indicate a good model fit because each of their values is above the minimum threshold of 0.900. In addition, the AOSR value of 0.034 and the RMSEA of 0.073 fall within their respective standards for acceptance. As Figure 2 illustrates, the resulting model is much more complex than the direct effects model. Role conflict now has a direct, positive influence on burnout (0.902) and an indirect effect through stress arousal. Role conflict also has an indirect negative influence on performance, through stress arousal. Role conflict also has a negative influence on job satisfaction (−1.436) and a positive influence on turnover intentions (1.712). Role ambiguity has a negative influence on performance (−0.178) and turnover intentions (−0.405). Although it decreases performance, as in the direct effects model, role ambiguity now appears to reduce turnover intentions. Counterintuitively, role overload is negatively related to burnout (−0.340) and turnover intentions (−0.932) and is positively related to job satisfaction (0.900) and performance (0.248). Overall, role conflict plays a major part and role overload an enhanced part in explaining the outcomes. Role ambiguity appears to play a reduced role. In turn, stress arousal has a significant positive influence on burnout (0.329) and a negative influence on performance (−0.345). Some changes in the direction or sign of the paths also are seen.
FIGURE 2 Stress Arousal as a Mediator Model Standardized Path Coefficients.
Paths between each latent construct and its indicators are omitted for ease of diagramming and interpretability (see Table 4 for these relations). *Covariance between independent factors (all significant at P < .01). **Significant at P < .01.
TABLE 5 Stress Arousal Model Goodness-of-Fit Test Results
With the exception of job satisfaction, the stress arousal model explains a larger increase in the variance for each of the outcomes than does the direct effects model (see Table 6).
TABLE 6 Comparison of Variance Explained for Each Key Outcome
The r 2 for burnout increased from 0.69 to 0.76 (10%). The r 2 for performance increased from 0.06 to 0.14 (133%). The r 2 for turnover intentions increased from 0.40 to 0.59 (47.5%). There was no change for job satisfaction (r 2 = 0.67). Thus, inclusion of stress arousal as a mediator increased the explanatory effects of role stressors on the job-related outcomes.
The emotional arousal captured by the stress arousal construct clearly is important to capturing the effects of role conflict. This becomes even clearer when looking at the total effects, which include the direct effect plus the product of the indirect effects through stress arousal. For role conflict to burnout, the total effects equal 0.90 + (0.67 × 0.33) = 1.12. Nineteen percent of the impact of role conflict on burnout is the result of the indirect effect. Role conflict has an indirect effect only on performance, through stress arousal. In this case the total effects equal 0 + (0.67 × −0.34) = −0.23. One hundred percent of role conflict's impact on performance is a mediated effect through stress arousal.
DISCUSSION
The potential influence of the role stressors examined in this study has been of considerable interest to accounting researchers through the years. As noted above, some have theorized that these are consequences of other environmental influences; however, another stream of research positions them as independent (ie, exogenous) predictors of stress and its consequences. Regardless, their influence as stress antecedents is well documented.
Previous research reports role conflict as having significant positive relations with stress arousalReference Smith, Davy and Everly21 and job tension,Reference Rebele and Michaels38Reference Fogarty39 burnout,Reference Smith, Davy and Everly21Reference Fogarty, Singh, Rhoads and Moore24Reference Sweeney and Summers40 and turnover intentions.Reference Viator41 Conversely, role conflict was found to have a negative relation to job satisfactionReference Smith, Davy and Everly21Reference Smith, Davy and Stewart32Reference Rebele and Michaels38 and performance.Reference Fogarty39 The present study supports these findings, but with a caveat. The relations among role conflict and burnout, turnover intentions, job satisfaction, and performance occur only when stress arousal is introduced as a mediator between the stressor and the outcomes. These results support the argument that stress arousing the cognitive-affective/emotional domain produces negative psychological and behavioral responses.Reference Lazarus and Folkman30Reference Everly and Sobelman31 The findings of this study go even further by demonstrating clearly that without the cognitive-affective response, role conflict has no effect on any of the outcomes.
The results for role ambiguity are not as clear. When examining the direct effects model, the results are consistent with previous research, with role ambiguity having positive relations with burnoutReference Smith, Davy and Everly21Reference Fogarty, Singh, Rhoads and Moore24 and turnover intentions.Reference Viator41 Role ambiguity is also negatively related to job satisfactionReference Rebele and Michaels38 and performance.Reference Viator41
When examining the mediator model, which includes stress arousal, the impact or involvement of role ambiguity changes considerably. Previous research shows role ambiguity having a significant positive relation to stress arousal in studies in which ambiguity was measured using items loading from a factor analysis of Kahn et alReference Kahn, Wolfe, Quinn and Snoek42 and Rizzo et alReference Rizzo, House and Lirtzman22 scales.Reference Smith, Everly and Johns33Reference Bentler and Wu34 The results from the present study do not show this relation. One possible explanation lies with the measures used. The findings of the present study are consistent with those of Smith et al,Reference Smith, Davy and Everly21 which also failed to find a significant relation between role ambiguity and stress arousal. Like us, they also used only Rizzo and colleagues’Reference Rizzo, House and Lirtzman22 items to measure the construct. The measures of Kahn et al were not used in either study. Role ambiguity also failed to show a significant relation with burnout and job satisfaction. Again, this may be the result of using different measures. It is possible that the measures of Kahn et alReference Kahn, Wolfe, Quinn and Snoek42 included some aspect(s) of role ambiguity not addressed by Rizzo et al.Reference Rizzo, House and Lirtzman22 This needs to be examined further.
As in the direct effects model, role ambiguity retained a significant negative relation with performance in the mediator model. Unlike the direct effects results, however, role ambiguity is negatively related to turnover intentions in the mediator model. The positive relation in the direct effects model is as one would intuitively expect.Reference Viator41 The findings in the mediator model are consistent with those of Smith et al,Reference Smith, Everly and Johns33 who reported a negative relation with turnover intentions. They provided 2 possible explanations for this counterintuitive finding: role ambiguity enhances an individual's insecurity,Reference Mackey and Cooper43 which in turn attenuates one's inclination to consider a job change, and individuals experiencing high role ambiguity do not perceive that alternative job opportunities offer lower levels of ambiguity. These do not explain why the sign of the path changed between the direct effects model and the mediator model.
The change in the sign and the loss of 2 significant paths suggests the inclusion of stress arousal. Although not directly related to role ambiguity, stress arousal does have an impact on how this stressor relates to outcomes. These findings can be explained by what is referred to as specification error in structural modeling.Reference James, Mulaik and Brett44 A construct that is a relevant cause (ie, moderately to highly correlated with other causes) that is not included in the model will lead to biased solutions. These biases may result in the wrong sign on paths to endogenous (outcome) variables and/or paths being deemed significant that are not and vice versa.Reference James, Mulaik and Brett44 As we can see from the results of role conflict and stress arousal, stress arousal is strongly related to role conflict. Thus, leaving stress arousal out of the direct effects model was a major specification error of the model. By including it in the model, the contribution of role ambiguity is better explained, as is that of role conflict.
We find similar problems when looking at the effects of role overload. In the direct effects model, role overload is positively related to turnover intentions and burnout. Previous research has also reported positive relations between role overload and these 2 constructs.Reference Smith, Davy and Everly21Reference Fogarty, Singh, Rhoads and Moore24Reference Sweeney and Summers40 Unlike previous studies, role overload was not related to job satisfaction or performance. In the mediator model, role overload is positively related to job satisfaction and performance and negatively related to burnout and turnover intentions.
Both Fogarty et alReference Fogarty, Singh, Rhoads and Moore24 and Smith et alReference Smith, Davy and Everly21 found significant positive relations between role overload and both job satisfaction and performance. In explaining these counterintuitive findings, Fogarty et alReference Fogarty, Singh, Rhoads and Moore24 proposed that overload includes an “eustress” component that is unmediated, and Smith et alReference Smith, Davy and Everly21 speculated that these relations may have resulted from individuals evaluating overload as a challenge rather than a threat, thus giving it the potential to promote personal gain and growth.Reference LePine, Podsakoff and LePine45 Similar to this study, Smith et alReference Smith, Davy and Everly21 included stress arousal as a predictor of job satisfaction and performance.
From a health perspective, role overload has been found to have a significant, positive direct relation to stress arousal.Reference Smith, Davy and Everly21Reference Smith, Everly and Johns33Reference Bentler and Wu34 That is not the case in the present study; role overload is not related to stress arousal. Moreover, Fogarty et alReference Fogarty, Singh, Rhoads and Moore24 found that overload had a significant positive relation to burnout, as did Sweeney and SummersReference Sweeney and Summers40 during their January through April “busy season” analyses. Although Smith and colleagues’Reference Smith, Davy and Everly21 initial replication of the Fogarty et al modelReference Fogarty, Singh, Rhoads and Moore24 also measured a significant positive direct relation between overload and burnout, this relation was no longer significant when stress arousal was added to the model as an antecedent to burnout, leading to the proposition that a direct path between the former 2 constructs may not be warranted. That conclusion appears a bit premature because this study shows a significant, negative relation between role overload and burnout with stress arousal included in the model. It is also negatively related to turnover intentions. These results are consistent with the positive relations with job satisfaction and performance. Thus, role overload may be seen as a positive challenge, as argued above. Alternatively, in today's economic conditions, role overload may be interpreted as job security, relieving another potential stressor.
Again, the change in the nature of the relation for role overload suggests serious model specifications problems with the direct effects model. Stress arousal plays a critical role in explaining the relations between the 3 stressors and the 4 outcome constructs examined in the present study.
CONCLUSIONS
This study used latent variable structural modeling to segregate the direct effects of adverse conditions vs the cognitive-affective mediating effects of those conditions upon selected job-related outcome within a resilience context. Our data showed that cognitive-affective mediation accounted for a 130% increase in explained variation in job performance, a 48% increase in variation in job turnover intention, and a 10% increase in burnout.
Our intention in conducting such an investigation was to attempt to identify a practical defining aspect of human resilience in the face of adverse conditions, with the subsequent intention of providing better guidance for the design and implementation of programs designed to enhance human resilience. The military, medicine, nursing, emergency services, and public health professions seem like professions that could prosper from enhanced resilience given their unusual exposure to adverse conditions. It may be that these data could serve to improve training programs for these “at risk” professional groups. The implications may be far broader in scope, however. Could these data assist in enhanced public health preparedness programs for the public? This seems to be a question worthy of additional investigation.
Author Disclosures: The authors report no conflicts of interest.
