Supervisory Communication, Burnout, and Turnover Intention Among Social Workers in Health Care Settings

Abstract

The current study tests the effects of different types of supervisory communication on burnout and turnover intention among health social workers. The study proposed a conceptual model of supervisory communication and tested it empirically using structural equation modeling (SEM) techniques with a random sample of 211 California state-registered social workers working in health or mental health care settings. The results of the present study provide empirical evidence of the unique roles that different types of supervisory communication play as antecedents of burnout and turnover intention. Specifically, supportive relationship communication had an indirect effect on burnout and turnover intention through its effect on perceived stress, whereas job-relevant communication had not only an indirect effect on burnout and turnover intention through its effect on stress, but also a direct effect on turnover intention. In addition, the results showed that upward communication moderated the relationship between stress and burnout. Implications for social work administration and possible elaboration of the theoretical framework are discussed.

KEYWORDS:

• supervisory communication
• burnout
• turnover intention
• social workers in health care

BACKGROUND

Social work delivers vital health services to a vast array of populations in need. Areas of practice within health social work include health promotion, general health care, HIV/AIDS (Hall, 2007), disease management (Claiborne & Vandenburgh, 2001), mental health care (Cohen, 2003), and health care with the elderly (Berkman, Gardner, Zodikoff, & Harootyan, 2005; Chong, 2007). In these fields, social workers play unique and essential roles that include discharging planning (Lechman & Duder, 2006) and resolving ethical dilemmas in hospice (Csikai, 2004). Given the current emphasis on health care cost control, productivity, and accountability, social workers' job attitudes and job performance in health care settings has become an important research topic.

In a recent report on the difficulties of the social work profession in general, examples of job demands included increasing paperwork and caseloads as well as staff shortages and reduced availability of adequate supervision (Center for Workforce Studies, 2006). Social workers in health care settings face even more demanding situations because of the increased role of managed care (Egan & Kadushin, 2004). Previous literature on burnout has suggested that demanding job conditions are significant antecedents of social worker burnout (Söderfeldt, Söderfeldt, & Warg, 1995). Social worker burnout is a critical managerial issue because social worker burnout negatively affects the quality, consistency, and quality of client services. A review of burnout literature suggests that worker burnout increases levels of depression (Glass & McKnight, 1996), sick-leave absences (Toppinen-Tanner, Kalimo, & Mutanen, 2005), turnover intention (Huang, Chuang, & Lin, 2003), actual turnover (De Croon, Sluiter, Blonk, Broersen, & Frings-Dresen, 2004), and self-reported health problems (Schaufeli & Enzmann, 1998). Although it is generally agreed that preventing burnout is important, only a few empirical studies have examined what alleviates burnout among social workers in health care settings.

Considering that social work is done in the context of supervisory relationships, effective supervision is essential to preventing burnout and turnover. Social work supervisors play the role of teacher, enabler, consultant, and manger for frontline social workers (Kadushin, 2002). An effective supervisor must provide a frontline social worker with essential guidelines, professional skills, and knowledge related to services while also communicating with social workers about opinions, feelings, and decision making with mutual trust and respect. This notion has been supported by studies reporting that perceived supervisory support, which is considered an outcome of supervisory communication (Ray & Miller, 1991), is associated with levels of burnout and turnover intention among those being supervised (Um & Harrison, 1998; Mor Barak, Nissly, & Levin, 2001).

Several explanations have been offered as to why supervisory communication helps prevent burnout and turnover. First, it reduces workers' role stress (i.e., role ambiguity and role conflict) (Miller, Ellis, Zook, & Lyles, 1990; Ray & Miller, 1991). Because role stress often leads to burnout or turnover (Cordes & Dougherty, 1993; Mor Barak, Nissly, & Levin, 2001), supervisory communication indirectly influences worker burnout and turnover through its effect on perceived role stress. Second, when an individual is experiencing a high level of role stress, supervisory communication is an important coping resource. Several studies have suggested that supervisory support buffers the effect of role stress on burnout (Bakker, Demerouti, & Euwema, 2005; Kickul & Posig, 2001). Despite the significant contribution of many studies, however, few have examined the different effects of supervisory communication on burnout and turnover intention among health care social workers in particular. The current study examines (1) the indirect effects of supervisory communication on burnout and turnover intention resulting from perceived role stress and (2) the moderating effects of supervisory communication on the relationship between role stress and burnout.

Supervisory communication can be described using certain characteristics, including the content of communication and the direction of communication flow (Krone, Jablin, & Putnam, 1987). This study focuses on three specific types of supervisory communication: job-relevant communication, upward communication, and positive relationship communication. A conceptual model hypothesizing the relationships among the three types of supervisory communication, role stress, burnout, and turnover intention is proposed and empirically examined through Structural Equation Modeling (SEM) techniques using data from a sample of California-registered social workers in health care settings.

LITERATURE REVIEW

Burnout and Turnover Intention

When it was first introduced, the term “burnout” referred to a phenomenon that was observed among human service workers who had to deal with emotionally demanding individuals. Since then, the majority of burnout studies have been based on Maslach and Jackson's (1986) conceptualization of burnout as having three components: emotional exhaustion (feelings of being overextended and depleted of emotional and physical resources), depersonalization or cynicism (negative or excessively detached responses to various aspects of the job), and diminished personal accomplishment (feelings of incompetence and a lack of achievement at work).

Theoretical frameworks of burnout (e.g., Cordes & Dougherty, 1993; Demerouti, Bakker, Nachreiner, & Schaufeli, 2001) have placed burnout as a key mediator of the relationship between chronic job stress and various strain outcomes. Research has consistently shown that a worker's level of emotional exhaustion is greatly affected by the nature and intensity of stress in the work environment (Cordes & Dougherty, 1993). A worker who is experiencing higher levels of emotional exhaustion is more likely to have depersonalized attitudes toward his or her clients and lack a sense of personal accomplishment at work (Cordes & Dougherty, 1993; Leiter & Maslach, 1988). This view of how burnout develops explains how job stress results in higher levels of not only emotional exhaustion but also depersonalization and diminished personal accomplishment. Comprehensive reviews of the burnout literature suggest that social workers are more likely to feel burned out when they perceive higher levels of job stress characterized as high role conflict, role ambiguity, and role overload (Söderfeldt, Söderfeldt, & Warg, 1995). Among strain outcomes is turnover intention, which several studies have empirically supported as a key outcome of burnout (Harrington, Bean, Pintello, & Mathews, 2001; Huang, Chuang, & Lin, 2003). Therefore, it is expected that role-related job stress (i.e., role stress), characterized as high levels of role conflict, role ambiguity, and role overload, is positively associated with burnout, and that burnout is positively associated with turnover intention among social workers.

Types of Supervisory Communication

Based on Miles, Patrick, and King's (1996) discussion, this study focuses on three major types of supervisory communication: job-relevant communication, upward communication, and positive relationship communication.

Job-relevant communication

Social work uses technologies that are not highly determinate and rely on the discretion of frontline workers (Patti, 2000). The general characteristics of social work amplify the importance of the supervisor's roles as a teacher or instructor for frontline social workers. Supervisory communication regarding job-related matters (i.e., job relevant communication) embraces performance feedback, information about rules and policies, work schedules and assignments, task-specific instructions, and goals (Miles, Patrick, & King, 1996). Previous studies have suggested that effective job-relevant communication between supervisees and their supervisors can increase professional knowledge or skill levels (Albrecht & Adelman, 1987), improve role definition, and raise feelings of accomplishment at work (Ellis & Miller, 1994) and job performance among supervisees (Kluger & DeNisi, 1996). In his national study of social work supervision, Kadushin (1992) also found that the most useful function of social work supervision was educational. Job-relevant communication with supervisors has even been found to be the key predictor of social workers' performance (York & Denton, 1990).

Upward communication

Upward communication is defined as information exchange in upward transmission (from lower to higher members in organizational hierarchy) (Roberts & O'Reilly, 1974). The literature on social work supervision has emphasized that the relationship between supervisor and social workers should be less hierarchical and more one of mutual influence (Hbert, 1992). When supervisors and frontline workers have mutual trust, respect, and rapport, the workers can more readily communicate their opinions and feelings upward. Upward communication from social workers to their supervisors is considered a key characteristic of an open communication relationship in which both parties perceive the other to be a willing and receptive listener (Jablin & Krone, 1994). For example, Bolon (1995) argued that successful supervisors rely on upward communication by creating a climate of trust and respect with supervisees, and Glauser (1984) observed that upward communication can help mangers and supervisors to recognize areas in need of change and adjustment. At the same time, upward communication helps workers to have higher levels of job satisfaction and work performance, so social workers who enjoy open communication relationships with supervisors will be more likely to perceive their supervisory relationships as participatory and trustworthy (Eisenberg & Witten, 1987).

Positive relationship communication

Positive relationship communication between a supervisor and a social worker occurs when supervisors interact with social workers in an informal and supportive manner. Supportive personal relationships lead workers to believe that they are valued by others, and having a supportive supervisor has been identified as an important condition for lowering levels of job stress. For example, evidence suggests that having a supportive supervisor is significantly related to higher levels of job satisfaction (Newsome & Pillari, 1991; Rauktis & Koeske, 1994) and that perceived supervisory support in the work place decreases the likelihood of worker burnout (Houkes, Janssen, de Jonge, & Bakker, 2003) and turnover intention (Mor Barak et al., 2001).

The Role of Supervisory Communication in Information Processing and Reducing Uncertainty

Supervisory communication serves a critical role in models of health social worker burnout in terms of reducing the perceived levels of job stress. Miller et al. (1990) discussed social information processing theory (Salancik & Pfeffer, 1978) and uncertainty reduction theory (Berger & Calabrese, 1975) as theoretical foundations with which to explain the role of supervisory communication in the process of burnout. First, the social information processing theory considers perceptions of the workplace to be a function of the communication environment in which workers are embedded, rather than a function of the objective characteristics of jobs and needs of workers (Salancik & Pfeffer, 1978). Beehr (1995) suggested that supervisor contact is likely to consist of instrumental support and affect the conditions that lead to job stress. Similarly, Leiter and Maslach (1988) argued that a major part of role stress comes from the extent to which workers perceive supervisors to be making irreconcilable demands on their time and resources. Thus, job-relevant communication between a supervisor and social worker is associated with social workers' perceived levels of role stress.

Upward communication and positive relationship communication is also related to social workers' role stress. Based on uncertainty reduction theory (Berger & Calabrese, 1975), Albrecht and Adelman (1987) discussed the role of social support in reducing uncertainty in organizational contexts. Schuler and Jackson (1986) stated that “the more participation allowed employees, the more likely they will be able to get a clear understanding of what is expected and what is rewarded. Consequently, the less uncertainty and stress there is for the individual” (as cited in Miller et al., 1990, 215). Miller et al. (1990) found that supervisory support reduced perceived role stress and the perceptions of workloads among permanent staff in hospitals. Therefore, the current study hypothesizes that social workers' attitudes and perceptions of the role-related stress, such as role ambiguity, role conflict, and role overload, will be a function of communication with the immediate supervisor.

Supervisory Communication as Empowerment and as a Job Resource

Models of job demands and resources (JD-R) (Demerouti et al., 2001) provide theoretical frameworks concerning why certain resource factors are critical in explaining the relationship between role stress and burnout. The central theme of the JD-R model of burnout is that “burnout develops when job demands are high and when job resources are limited because such negative working conditions lead to energy depletion and undermine employee's motivation, respectively” (Demerouti et al., 2001, 499). Job resources are defined as physiological, psychological, social, or organizational aspects of the job that (a) are “functional in achieving work goals, (b) reduce job demands and the associated physiological and psychological costs, or (c) stimulate personal growth and development” (Demerouti et al., 2001, 501). According to the JD-R model, when the working environment lacks resources, social workers cannot cope with the negative influences of role stress. Recently, Bakker et al. (2005) reported that job demands (e.g., work overloads, emotional demands, unfavorable working conditions, and work–home interference) had a weaker or no relationship with burnout in cases when job resources (e.g., social support, high quality of relationship with the supervisor, and performance feedback) were available in the working environment. Based on the JD-R model of burnout, this study considered effective supervisory communication as job condition that not only reduces perceived role stress, but is also helpful in coping with perceived role stress.

Empowerment theory is another theoretical framework that explains the moderating effects of supervisory communication on the relationship between role stress and burnout. Conger and Kanungo (1988) suggested that supervisory practices that include expressing confidence in subordinates, fostering opportunities for subordinates to participate in decision making, and providing autonomy from bureaucratic constraint are more likely to be empowering. This study considers intrinsically rewarding job characteristics such as open communication between social workers and their supervisors as key elements of empowerment strategies and practices for social workers. Open communication with immediate supervisors allow social workers to access resources, information, and support in the organization to get the job done successfully; in other words, enhancing the motivating potential of jobs by allowing frontline social workers to participate in decision making empowers social workers and intensifies affective reactions toward the job. Upward communication, job-relevant communication, and supportive relationship communication can be also empowerment factors for social workers. For example, Eby, Freeman, Rush, and Lance (1999) used SEM based on meta-analytic correlations to identify performance feedback and supervisory satisfaction as key intrinsic motivators. Therefore, the current study hypothesizes that social workers with high quality of supervisory communication in terms of job-relevant communication, upward communication, and positive relationship communication will be less likely to burn out, compared to social workers who do not.

A Proposed Hypothetical Model

Based on the literature review, the following relationships between supervisory communication, role stress, burnout, and turnover intention are hypothesized:

• Hypothesis 1: Role stress will be positively associated with burnout in social workers.

• Hypothesis 2: Burnout will be positively associated with turnover intention in social workers.

• Hypothesis 3: Supervisory communication (i.e., job-relevant communication: H3-a, upward communication: H3-b, positive relationship communication: H3-c) will be negatively associated with role stress in social workers.

• Hypothesis 4: Supervisory communication (i.e., job-relevant communication: H4-a, upward communication: H4-b, positive relationship communication: H4-c) will moderate the relationship between role stress and burnout in social workers.

To test these hypotheses, a hypothetical model was developed. Figure 1 illustrates the hypothesized relationships between key constructs.

FIGURE 1 A proposed conceptual model of supervisory communication, role stress, burnout, and turnover intention.

Several studies have identified the importance of demographic factors in explaining worker burnout and turnover intention. For example, Brewer and Shapard's (2004) meta-analytic review of burnout studies reports age as negatively correlated to worker burnout, and Jackson (1993) found that gender is a significant demographic factor of burnout (Brewer & Shapard, 2004). Other studies have suggested that workers with shorter organizational tenure (Somers, 1996) and higher dissatisfaction with salary (Currall, Towler, Judge, & Kohn, 2005) are more likely to quit their jobs. Therefore, in order to develop a more valid and complete model of turnover intention, age, gender, organizational tenure, and annual salary have been included in the hypothetical model as control variables.

METHODOLOGY

Sample and Procedure

For the study's cross-sectional survey design, 1500 registered social workers were randomly selected from the total population of state-registered social workers working in California (n = 21,518) in 2005. Each of these 1500 social workers was mailed a survey instrument, an introductory letter, an information sheet, and a return postage-paid envelope. Of the 529 questionnaires returned, 51 were ineligible because respondents were currently retired or working in other fields (43), registered after the data-collection period (5), or failed to answer a significant number of questions (3). From the original sample of 1500, then, 478 questionnaires were eligible, yielding a survey response rate of 32%. Among eligible participants, 405 social workers were working full time or part time in organizational settings. A sub-sample of 211 full-time social workers in health care or mental health care settings was also used.

Measures

Supervisory communication

Communication with one's supervisor was measured by a short form (24-item version) of Huseman, Hatfield, Boulton, and Gatewood's (1980) original 56-item instrument representing eight message types: feedback, rationale, information, direction, negative expression, positive expression, participation, and upward openness. Each question was measured with a five-point Likert-type scale. Miles, Patrick, and King (1996) performed exploratory factor analysis for this instrument and identified four types of supervisory communication: job-relevant communication, upward communication, positive relationship communication, and negative relationship communication. In the present study, three items of negative relationship communication were excluded because social workers (n = 48) who participated in a pilot study suggested that these items were not appropriate to research in social work. These three items are “My supervisor ridicules or makes fun of me,” “My supervisor criticizes my work in front of others,” and “My supervisor is critical of me as a person.” After omitting these items, a principal component factoring analysis with varimax rotation was performed for the modified 21-item version of Huseman et al.'s measure. Results suggested three sub-factors (i.e., job-relevant communication, upward communication, positive relationship communication) structure of the measure, which were consistent with Miles et al.'s 1996 study. Job-relevant communication consisted of ten items (e.g., “My supervisor gives clear instructions to me” and “My supervisor lets me know why changes are made in work assignments”). Positive relationship communication consisted of seven items (e.g., “My supervisor strikes up casual conversations with me”). Upward communication consisted of three items (e.g., “I question my supervisor's instructions when I don't understand them” and “I tell my supervisor when I think things are being done wrong”). Cronbach's alphas for this study sample were 0.95 for the job-relevant communication scale, 0.92 for the positive relationship communication scale, and 0.77 for the upward communication scale.

Role stress

Role stress was assessed by measuring three role-related stressors: role conflict, role ambiguity, and role overload. Role conflict (RC) and role ambiguity (RA) were measured using a shortened form of the Rizzo, House, and Lirtzman (1970) role conflict/role ambiguity questionnaire. The role conflict scale consisted of eight items (e.g., “I receive an assignment without the manpower to complete it”) and included questions regarding: (1) conflict between the respondent's internal standards and the defined role behavior; (2) conflict between the time, resources, or capabilities of the respondent and the defined role behavior; (3) conflict between several roles for the same person that require different behaviors; and (4) conflict between expectations and organizational demands in the form of conflicting requests with incompatible standards of evaluation. Role ambiguity consisted of six items (e.g., “I know what my responsibilities are”). Participants were asked to respond to the six-item RA and eight-item RC scales by indicating on a 7-point scale ranging from “very false” (1) to “very true” (7) the degree to which the condition applied to them. Cronbach's alpha for the internal consistency and reliability for this sample was 0.88 for the RC scale and 0.84 for the RA scale. Role overload was measured by three items from the five-item scale of work overload, which were specifically designed to measure workload among human service workers (Lait & Wallace, 2002): (1) “I have to attend too many meetings in this job,” (2) “My job involves a lot of paperwork,” and (3) I have to work very fast to get everything done in my job.” This scale also employed a 7-point Likert-type scale ranging from “strongly disagree” (1) to “strongly agree” (7). The Cronbach's alpha for its internal consistency and reliability was 0.60.

Burnout

Burnout among social workers was assessed using Maslach's Burnout Inventory–Human Service Survey (MBI–HSS; Maslach & Jackson, 1986), which measures burnout components by asking about the frequency with which workers experience feelings related to each aspect of the burnout syndrome. The survey included nine questions on perceived emotional exhaustion, five questions on depersonalization, and eight questions on personal accomplishment (Maslach & Jackson, 1986). Each statement was rated on a 7-point continuum from “never” experienced (0) to experienced “every day” (6). For a sample of social service workers, the MBI showed internal reliability (Cronbach's alpha ranging from 0.71 to 0.90) and test–retest reliability (2- to 4-week intervals for all scales, ranging from 0.60 to 0.82) (Maslach & Jackson, 1986). Cronbach's alpha for the current study was 0.91 for emotional exhaustion, 0.75 for depersonalization, and 0.79 for personal accomplishment.

Turnover intention

Organizational turnover intention was measured by three items from the four-item scale of intention to leave (Nissly, Mor Barak, & Levin, 2005): (1) “In the next few months, I intend to leave this organization,” (2) “In the next few years, I intend to leave this organization,” and (3) “I occasionally think about leaving this organization.” These items were rated on a 7-point Likert-type scale ranging from “strongly disagree” (1) to “strongly agree” (7), for which the Cronbach's alpha was 0.76.

Control variables

The questionnaire included questions on respondents' age, gender, organizational tenure, and annual salary. Organizational tenure was determined by the number of years the respondent has worked for his or her current employer, and annual salary was based on respondents' total yearly income from their current jobs.

Analysis

Missing values for all items in the standardized scales (e.g., MBI–HSS) were replaced by plausible values based on the multiple imputation method using the Windows freeware, NORM (Schafer, 1999). Following the method described by Olsen and Schafer (1998), we created three imputed data sets and obtained a single-point estimate by averaging across the estimates from the data sets. However, demographic variables still have missing values: 1.6% of the samples were missing an age, 1.9% were missing organizational tenure, and 3.1% were missing annual wage. In order to handle these missing values with no loss of information in the available dataset, we used the Full Information Maximum Likelihood (FIML) Method (Arbuckle, 1996). In order to use the FIML method, the data must be multivariate normal. Skewness ranged between − 0.30 and 1.46, and kurtosis ranged between − 0.96 and 1.77, so practically normal distributions were assumed for all variables.

After the preliminary step, structural equation analyses using the Mplus 4.0 program (Muthén & Muthén, 2006) were conducted with a FIML method. First, composite scores for three types of supervisory communication, three kinds of role stressors, three dimensions of burnout, and turnover intention were calculated. Second, as suggested by Anderson and Gerbing (1988), a confirmatory factor analysis (CFA) was performed to test a measurement model that included two latent variables, role stress and burnout. Once the measurement model was estimated and convergent, and proper solution was derived, a third step assessed a series of structural equation models based on the proposed conceptual model. To evaluate model fit, the χ²/df, comparative fit index (CFI), and the Root-Mean-Square Error of Approximation (RMSEA) were examined. A value of χ²/df less than 3 indicated a reasonable fit (Kline, 1998). CFI can range from 0 to 1, where 0 represents the goodness of fit associated with the null model and 1 represents the goodness of fit associated with a saturated model; a value above 0.90 suggests an acceptable fit between the model and data (Kline, 1998). Finally, the RMSEA is the most sensitive index for models with misspecified factor pattern coefficients and the least sensitive to the sample size compared to other model fit indexes. An RMSEA of 0.05 or less indicates a good fit and an RMSEA of 0.08 or less indicates a reasonable error of approximation (Browne & Cudeck, 1993).

In the final step, multi-group analyses were performed to test the moderating effects of supervisory communication in this model. The main question of Hypothesis 4 was whether supervisory communication moderates the relationship between role stress and burnout in social workers. Another way of expressing the question in terms of an interaction effect is whether group membership (i.e., high supervisory communication vs. low supervisory communication) moderates the relationship between role stress and burnout. For this analysis, the sample was divided into two sub-samples based on whether the respondent was above or below the median on each type of supervisory communication. Three estimated models were used to test the moderating effect on each path (between role stress and burnout). In the first model, Model 1, we freely estimated the models for high and low on each type of supervisory communication. In Model 2, all of the factor loadings were constrained to be equal. In Model 3, a path between role stress and burnout was constrained to be equal across groups. A non-significant chi-square difference between Model 1 and Model 2 would indicate measurement invariance across groups. If factor loadings were equivalent across groups, the moderating effects could be examined by examining chi-square difference between Model 2 and Model 3. A significant chi-square difference between Model 2 and Model 3 would suggest that the equality constraints are not consistent with the data and, thus, that an interaction effect exists (Rigdon, Schumacker, & Wothke, 1998).

RESULTS

Table 1 presents the means, standard deviations, and correlation coefficients for 14 observed variables in the final structural equation model. The results of the confirmatory factor analysis, including two latent variables, did not support the measurement model. Overall, there was a χ² (8) value of 27.48, with χ²/df = 3.44, CFI = 0.93, and RMSEA = 0.107. A standardized factor loading for personal accomplishment was small (β = − 0.24), and the burnout factor only marginally explained variance in personal accomplishment (R² = 0.06). It is consistent with previous findings that emotional exhaustion and depersonalization may be key components reflecting a high degree of burnout among social workers (Walkey & Green, 1992). Therefore, personal accomplishment was excluded from the measurement model. Because the desired number of indicators for a latent variable is at least three (Kline, 1998), items of emotional exhaustion were combined to form two item parcels so that three measurement indicators (i.e., emotional exhaustion 1, emotional exhaustion 2, and depersonalization) represented the latent construct of burnout. The new measurement model yielded an overall χ² (8) value of 19.80, with χ²/df = 2.48, CFI = 0.98, and RMSEA = 0.084 (PCLOSE = 0.103). Although the value of RMSEA was larger than 0.08, the probability of close fit was 0.103. Because the measurement model was found to be a proper solution for the data, structural paths were specified among these latent constructs, and other variables and structural models were examined.

TABLE 1 Means, Standard Deviations, and Correlation Coefficients for Observed Variables

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. JC	1.00
2. PC	0.75∗	1.00
3. UC	0.35∗	0.48∗	1.00
4. RC	− 0.31∗	− 0.29∗	0.01	1.00
5. RA	− 0.40∗	− 0.29∗	− 0.08	0.42∗	1.00
6. RO	− 0.27∗	− 0.24∗	− 0.06	0.52∗	0.29∗	1.00
7. EE1	− 0.26∗	− 0.31∗	− 0.11	0.47∗	0.37∗	0.49∗	1.00
8. EE2	− 0.25∗	− 0.29∗	− 0.07	0.51∗	0.33∗	0.54∗	0.90∗	1.00
9. DP	− 0.05	− 0.08	− 0.03	0.34∗	0.21∗	0.23∗	0.51∗	0.49∗	1.00
10. TI	− 0.37∗	− 0.36∗	− 0.07	0.27∗	0.28∗	0.24∗	0.39∗	0.39∗	0.06	1.00
11. Age	− 0.08	− 0.12	− 0.06	0.09	0.00	0.00	0.02	0.01	− 0.08	0.10	1.00
12. Gender	− 0.05	− 0.07	− 0.08	− 0.15∗	− 0.08	− 0.10	− 0.04	− 0.04	− 0.13	− 0.02	− 0.25∗	1.00
13. Tenure	− 0.05	− 0.07	0.02	0.02	− 0.12	− 0.01	0.01	0.00	− 0.09	0.08	0.57∗	− 0.18∗	1.00
14. Salary	0.08	0.05	0.06	0.08	− 0.14∗	0.04	− 0.02	0.05	0.06	− 0.09	0.33∗	− 0.29∗	0.42∗	1.00
Mean	3.48	3.46	4.07	4.35	2.77	5.11	3.11	2.90	1.47	3.83	45.11	0.78	7.99	57.47
SD	1.06	1.09	0.68	1.45	1.05	1.19	1.36	1.28	1.17	1.73	12.00	0.41	7.53	17.91
JC: job relevant communication; PC: positive relationship communication; UC: upward communication; RC: role conflict; RA: role ambiguity; RO: role overload; EE1: emotional exhaustion 1; EE2: emotional exhaustion 2; DP: depersonalization; TI: turnover intention; Tenure: organizational tenure; Salary: annual salary.
∗p < .05 (2-tailed).

The proposed path model yielded an overall χ² (45) value of 100.91, with χ²/df = 2.24, CFI = 0.93, and RMSEA = 0.074 (PCLOSE = 0.026). Because a value of PCLOSE suggested that the model fit was marginal, we examined a residual covariance matrix and modification index. The small and near-zero residuals suggested that the model fit the data well. Some of the standardized residuals for turnover intention were larger than 0.20, which suggested that some of relationships with turnover intention were misspecified. Furthermore, the modification index suggested that an error correlation between job-relevant communication and turnover intention was significant. Therefore, a direct path from job-relevant communication to turnover intention was also specified. The modified model yielded an overall χ² (46) value of 84.06, with χ²/df = 1.83, CFI = 0.95, and RMSEA = 0.063 (PCLOSE = 0.157). For the purpose of achieving a more parsimonious model, the proposed model was further modified by trimming non-significant paths, based on the series of chi-square significant tests, but the non-significant path was removed from the model only if removing it did not influence the model fit significantly. As a result, the final model (Figure 2) yielded an overall χ² (54) value of 89.21, with χ²/df = 1.65, CFI = 0.95, and RMSEA = 0.056 (PCLOSE = 0.310). In this final model, the squared multiple correlation for turnover intention indicated that, once the effects of age, gender, organizational tenure, and annual salary were accounted for, the model explained 51% of variances in burnout and 26% of variance in turnover intention among social workers.

FIGURE 2 Standardized coefficients for the structural equation model. Only significant paths at the levels of p < .05 were included in this figure after accounting for effects of age, gender, organizational tenure, and annual salary on role stress, burnout, and turnover intention. EE1: emotional exhaustion item parcel 1; EE2: emotional exhaustion item parcel 2; DP: depersonalization.

Figure 2 reports all standardized path coefficients and illustrates the relationships among constructs in the model. Specifically, job relevant communication and positive relationship communication with immediate supervisor were negatively associated with role stress. In addition, role stress had an indirect effect on turnover intention through burnout. Role stress was strongly associated with burnout (β = 0.72), but not directly with turnover intention. Burnout was significantly associated with turnover intention among social workers (β = 0.34).

Regarding the hypothesized moderating effects, a series of multi-group analyses was performed. The equivalency of factor pattern coefficients across groups was first examined by examining the chi-square difference between Model 1 and Model 2. With four degrees of freedom, the chi-square difference between Model 1 and Model 2 was significant for job-relevant communication, but factor loadings of two latent variables were invariant across groups of positive relationship communication (high positive relationship communication group vs. low positive relationship communication group) or upward communication (high upward communication group vs. low upward communication group). Therefore, in Model 3, the path from role stress to burnout was equally constrained across groups of positive relationship communication or upward communication. With one degree of freedom, a significant chi-square difference for upward communication indicated that upward communication had a significant moderating effect on the relationship between role stress and burnout (see Table 2).

TABLE 2 Results of Chi-Square Difference Tests for the Hypothesized Moderating Effects of Supervisory Communication

	Types of supervisory communication
Models	Job relevant communication Low: n = 101 High: n = 110	Positive relationship communication Low: n = 104 High: n = 107	Upward communication Low: n = 129 High: n = 82
Model 1	83.90 (df = 62)	92.20 (df = 62)	103.49 (df = 62)
Model 2	103.57 (df = 66)	98.26 (df = 66)	108.01 (df = 66)
Model 3	—	98.92 (df = 67)	111.87 (df = 67)
Δχ^2 (Δ4) between M1 & M2	19.67∗	6.06	4.52
Δχ^2 (Δ1) between M2 & M3	NA	0.66	3.86∗
Model 1: no constraints; Model 2: equal constraints on four factor loadings; Model 3: equal constraint on the path between role stress and burnout.
∗p < .05 (2-tailed).

DISCUSSION

The present study provides empirical evidence of distinctive effects of different types of supervisory communication on burnout and turnover intention among social workers. In particular, job-relevant communication and positive relationship communication have a negative correlation with burnout resulting from perceived role stress in health social workers. It appears that job-relevant communication and positive relationship communication are closely related with each other, but each type of communication has unique effects on social workers' perceptions of role stress. This supports Harkness's (1995) finding that supervisory skills and relationships have independent helping effects in supervised practice. Effective social work supervision requires not only professional knowledge of practice, but also skills in coordinating work and workload, setting limits and manageable goals, and monitoring work process for frontline social workers (Kadushin, 2002). In order for the supervision to be effective, it must include a climate of belief and trust. Job-relevant communication will be more effective if a supervisor and a worker interact with each other in supportive manners. According to Jain's (1973) study on the relationship between the effectiveness of supervisory communication and supervisors' performance in hospitals, workers perceive as satisfactory the information they receive regarding hospital policies and tasks if they perceive that the communication from the supervisor as generally positive and helpful. Jain also addressed that effective supervisory communication encouraged workers to exchange a great deal of information about task-related matters. In this vein, effective supervisory communication allows supervisors to understand social workers' opinions about their current working experiences, like workload or role conflict, based on direct input from the frontline social workers. If social workers express their role as unmanageable and stressful, supervisors could facilitate information-sharing up and down the hierarchy to help administrators understand how management processes make it difficult for frontline workers to commit to their jobs.

The current study found that job-relevant supervisory communication had significant direct effects on turnover intention among social workers in health care settings. Previously, Cohen and Gagin (2005) found that staff development actually reduces burnout among hospital social workers. The current findings suggest that professional development through the supervised practical experience is not only negatively associated with social workers' burnout but is also positively associated with worker attachment to the organization. However, the effect of job-relevant supervisory communication on turnover intention may be more salient among social workers in the early stage of developmental process. Shechtman and Wirzberger (1999) concluded that workers differ in their expectations of supervision according to their professional stage of development. Similarly, Thomas (1982) presented stages of professional development in child care workers and suggested that the new worker is primarily concerned with survival through knowledge of agency rules and service policies and is especially dependent on the supervisor as the major source of knowledge. Considering that social workers are more likely to experience burnout during the first two years (Cherniss, 1980), the future study needs to focus on understanding effective supervisory skills for social workers at the beginning of their professional careers.

The present study also found that upward communication buffers the effect of role stress on burnout. This suggests that, although social workers perceive high levels of role stress, they are less likely to experience burnout if they can discuss role stress with their immediate supervisors. Previous studies have focused primarily on supervisory support (Posig & Kickul, 2003) or supervisory performance feedback (Bakker, Demerouti, & Euwema, 2005) as stress-buffering factors. The current findings clarify that open upward communication is the kind of supervisory communication that actually buffers the effect of role stress on burnout.

LIMITATIONS AND RECOMMENDATIONS FOR FUTURE STUDY

The current study has certain limitations, particularly the methodological limitations of a cross-sectional design in assessing structural models that include mediation. For future research, researchers should employ a longitudinal design to provide empirical information about the causal relationships. In addition, it is unclear how well these findings could be generalized to social workers in health care settings because the sample was limited to registered social workers in California. Future studies should reexamine the conceptual model using a more representative sample. Such studies would increase understanding of the functions of supervisory communication in preventing burnout and turnover in social work in health care settings.

The current study's sample consisted of California registered social workers in health care settings, about 37% (n = 78) of whom were Associate Social Workers (ASWs). In California, an ASW must meet certain requirements, including 3200 supervised clinical hours to be a Licensed Clinical Social Worker (LCSW) (NASW California Chapter, 2007). During the licensure, a supervisor plays critical roles as a teacher and educator to help ASWs learn how to make clinical judgments and how to use theory according to the particular situations. To advance empirical knowledge on social work supervision, future research should examine whether effects of supervisory communication on burnout or turnover intention are different depending on stages of professional development among social workers.

Future study should also expand the current model by including other moderating effects of supervisory communication. For example, Um and Harrison (1998) reported that social support moderated the relationship between burnout and job dissatisfaction among social workers, and Kim and Stoner (2008) found that, although social workers perceive high levels of role stress, a supportive working environment helps them maintain their attachment to the organization. A more comprehensive model of supervisory communication should include the moderating effects of supervisory communication on the relationship between burnout and turnover intention, as well as the relationship between role stress and turnover intention.

CONCLUSION

The major implication of this research is that developing or improving supervision is a key strategy for preventing burnout and turnover in health care settings. Many studies have suggested that jobs should be redesigned so the levels of job autonomy and social support are increased in an effort to prevent burnout and retain workers (Dollard, Winefield, Winefield, & de Jonge, 2000; Kim & Stoner, 2008). High-quality supervision, as measured by effective communication practices, can lead social workers to perceive their working conditions as decentralized and supportive. Therefore, human-service managers should focus on developing organizational support, not only for frontline social workers, but also for supervisors because their skills and relationships with frontline workers are critical assets and resources for the organization.

This research was supported in part by a research award from the Hamovitch Center for Science in the Human Services and by an Albert and Frances Feldman Endowed Fellowship from the USC School of Social Work.