ABSTRACT
Research was conducted on the predictors of burnout in a sample of teachers in Queensland private schools. A total of 246 teachers responded to scales that assessed burnout, school and classroom environments, work pressure, role overload, role ambiguity, role conflict, teaching efficacy, external locus of control, and self-esteem. The Maslach Burnout Inventory was used to assess three facets of
burnout: emotional exhaustion, depersonalisation and personal accomplishment. An hypothesized model of burnout was tested in a LISREL analysis with post hoc modifications indicating that role overload, work pressure, classroom environment and self-esteem were predictors of emotional exhaustion. Depersonalisation was significantly related to emotional exhaustion, role conflict, selfesteem
and school environment. Teaching efficacy, self-esteem and depersonalisation were predictors of personal accomplishment.

In 1974, Freudenberger introduced the term burnout to describe the inability to function effectively in one's job as a consequence of prolonged and extensive job related stress. Since that time, incidences of, and research into stress and burnout have increased with popular emphasis on employees in the human services sector including social workers, nurses, teachers, lawyers, medical doctors and police officers (Jackson, Schwab, & Schuler, 1986; Maslach & Jackson, 1981). A common characteristic of these occupations is that the nature of the work can be highly emotional. For teachers, the potential for emotional stress is high since they work with classes of up to 35 students for long periods of time. The intensely relational nature of classrooms means that teachers are vulnerable to emotionally draining and discouraging experiences (Maslach & Leiter, 1999). Such experiences can lead to dysfunctional teacher behaviour with obvious implications for the teacher’s well-being and student learning.

This article reports the findings of a study of burnout in Queensland private school teachers. Specifically, the study investigated the influence of several hypothesized predictor variables. To provide a contextual basis for the research, background information on theoretical and empirical perspectives relating to this research is provided.

THEORETICAL AND EMPIRICAL PERSPECTIVES ON BURNOUT
According to Byrne (1991) and Maslach, Jackson, & Leiter (1996), the burnout syndrome has three distinct but loosely coupled dimensions: emotional exhaustion (feelings of being emotionally overextended and exhausted with one's work), depersonalisation (the
development of negative and uncaring attitudes towards others), and negative personal accomplishment (the loss of feelings of self-competence and dissatisfaction with one's achievements). Maslach et al. have developed and validated the Maslach Burnout Inventory
(MBI), an instrument that assesses these three dimensions. This instrument has been used in burnout research across a wide range of human environments.
Australian and overseas research has shown that high school teachers exhibit high levels of stress when compared to other white collar workers (Bransgrove, 1994). Otto (1986) showed that as many as one third of Australian teachers reported being very or extremely stressed. Teachers operating under high levels of stress for significant periods of time can develop burnout characteristics including less sympathy towards students, reduced tolerance of students, failure to prepare lessons adequately and a lack of commitment to the teaching profession. It follows that the study of teacher burnout is of great importance to the productivity of teachers and subsequent student learning.
Early attempts to describe stress and burnout emphasized their personal nature and, accordingly, blamed the individual teacher. This view has been superseded by a more social view of burnout that recognizes both background personality variables of the individual and
school characteristics as contributing to burnout in teachers. However, most studies of burnout have focused largely on the investigation of background variables like marital status, age, years of teaching and gender as predictors of burnout (Anderson & Iwanicki, 1984;
Byrne, 1991, 1994; Malik, Mueller, & Meinke, 1991; Maslach & Jackson, 1981). In fact, empirical studies involving psychosocial environment dimensions of schools and classrooms as antecedents to teacher burnout are rare.
According to Guglielmi and Tatrow (1998), serious conceptual problems have confronted stress and burnout research. Two examples demonstrate the divergent findings that can arise if variables are operationalized in quite different ways. On the influence of student
misbehaviour on teacher stress, Hart, Wearing and Conn (1995) concluded that ‘there is little point in trying to reduce teacher stress by reducing student misbehaviour’ (p. 27). By contrast, Boyle, Borg, Falzon and Baglioni (1995) reported that workload and student misbehaviour accounted for the most variance in predicting teaching stress. Hart et al. measured student misbehaviour with a single self-report item that assessed the time that the teacher spent dealing with student misbehaviour. It could be argued that such a simplistic and naïve conceptualisation of student misbehaviour does not in any way reflect the complex student misbehaviour issues that teachers handle on a daily basis and which are not related to time.
Similarly, the measure of organizational climate employed by Hart et al. is simplistic and does not reflect advances in school climate research since the early 1980s (see, e.g. Fraser, 1994). It seems clear that different researchers operationalize constructs in quite different ways. Recent research involving burnout has investigated links between teacher burnout and perceived self-efficacy in classroom management (Brouwers & Tomic, 2000), compared stress and burnout in rural and urban schools (Abel & Sewell, 1999), and studied the sources of stress and burnout in Hong Kong teachers (Tang & Yeung, 1999). Although research on learning environments and teacher burnout have shown remarkable progress over the past 25 years, no studies utilizing the latest approaches to research in these two fields have been conducted. The recognition of school and classroom environments as possible predictors of burnout is consistent with Lens’s and Jesus’ (1999) psychosocial interpretation of teacher stress and burnout and Maslach’s (1999) view that the social environment is at the heart of both understanding the teacher burnout phenomenon and ameliorating it.

Design of Present Study
The aims of the present study were to:
• validate scales to assess possible predictors of teacher burnout (viz. school and classroom environment, work pressure, role overload, role conflict, role ambiguity, teaching efficacy, locus of control, self-esteem) and Maslach Burnout Inventory scales (viz. emotional exhaustion, depersonalisation, and personal accomplishment), and
• investigate whether the postulated model of relationships among the above predictors and Maslach Burnout Inventory scales shown in Figure 1 fits the data through the use of structural equation modelling.

Figure 1. Postulated structural model for teacher burnout (observed variables,
fixed path loadings from observed variables to latent variables and error
variances for observed variables have been omitted.)

As shown in Figure 1, both organizational and personality variables predict burnout variables. It is noteworthy that this model was based on Byrne’s (1994) research that has related a host of variables with the three scales of the Maslach Burnout Inventory.

METHOD
Participants
The sample employed in this study consisted of 246 teachers who teach in private (i.e. nongovernment) schools in Queensland. Table 1 describes the sample which consisted of 99 primary, 103 secondary and 44 teachers from combined primary and secondary schools. As
indicated earlier in this article, age and gender have been shown to influence teacher burnout. While Table 1 describes the sample in terms of gender and age group, these two variables are not the focus of the present investigation.
Instrumentation
A test battery consisting of several instruments was administered to each respondent. All instruments had been employed in previous research in the United States but it was considered mandatory that the psychometric properties of each scale be reported. Details of
the specific instruments which are described in Table 2 are as follows:

• Classroom environment. A 24-item instrument which assesses teacher’s perceptions of their classroom psychosocial environments was used. Items were taken from four scales of a contemporary classroom environment instrument, the What is Happening in this Classroom (Aldridge & Fraser, 2000; Fraser, 1998). These scales assessed Interactions, Cooperation, Task Orientation, and Order and Organization in the classroom. Because of the problematic nature of conducting structural equation modelling with a large number of observed variables, a single classroom environment score based on a linear combination of item responses using
  factor scores as coefficients was computed and used in subsequent modelling. These factor scores were obtained from a confirmatory factor analysis (CFA). All classroom environment items employed a 5-point response format (viz. Strongly Disagree, Disagree, Not Sure, Agree, and Strongly Agree).
• School environment. In an analogous manner to the assessment of classroom environment, school-level environment, was assessed with 36 items from an instrument employed previously in school environment research (Dorman, Fraser, & McRobbie, 1997).
  These items were from six underlying scales (viz. mission consensus, empowerment, student support, affiliation, professional interest, and resource adequacy). As with classroom environment, a single school environment score based on a linear combination of item responses using factor scores from a CFA as coefficients was computed. The response format for all school environment items was the same as for classroom environment items.
• Role conflict, role ambiguity and role overload. Three 5-item scales that have been validated in previous research by Pettegrew and Wolf (1982) were used. These scales have been successfully used by Byrne (1994) in teacher burnout research in North America. All items employed a five point response format (viz. Strongly Disagree, Disagree, Not Sure, Agree, and Strongly Agree).
  

Table 2: Descriptive Information for Nine Predictor and Three Burnout Scales

• Teaching efficacy. This 7-item scale is from the Patterns of Adaptive Learning Survey(PALS) (Midgley et al., 1997). Items employed the same five point response format used for the above scales.
• Self-esteem. Seven items from the adult form of Coopersmith’s (1981) Self-Esteem Inventory were used. A five-point scoring format was used (viz. Very Unlike Me, Unlike Me,Neither, Like Me, Very Like Me).
• External locus of control. Byrne (1994) suggested the use of Rotter's Locus of Control scale (MacDonald, 1974; Rotter, 1966) as a measure of locus of control in burnout research. A modified 10-item form of Rotter's Locus of Control scale was used. Items relating
  to internal locus of control were reverse-scored so that scale scores were an indication of the respondent's perceived level of external locus of control. Items used a five point response format: 1 (Strongly Disagree), 2 (Disagree), 3 (Not Sure), 4 (Agree), and 5 (Strongly Agree).
• Burnout. A set of 19 items from the latest version of the Maslach Burnout Inventory Form ES (MBI) (Maslach et al., 1996) which has been developed especially for educational institutions was used to provide a self-assessment of each teacher's perceived burnout level. The original 22-item MBI has three factor-analytically derived scales: emotional exhaustion, depersonalisation and personal accomplishment. Whereas emotional exhaustion and depersonalisation are positively related to burnout, personal accomplishment is negatively related to burnout. A five-point Likert response format ranging from Almost Never to Almost Always was used to score each item.

Data Analysis and Interpretation
To investigate relationships among the above variables, structural equation modelling (SEM) using LISREL 8.3 (Joreskog & Sorbom,1993) was conducted. A weighted least squares (WLS) method with data from polychoric correlation and asymptotic covariance matrices was
used in the analyses. The WLS method was preferred because item data had five response categories, and polychoric correlations rather than Pearson product.moment correlations were computed. In these circumstances, Joreskog and Sorbom (1993) have argued that WLS is the appropriate method of analysis.
There were two distinct components to the analyses conducted in the present study. First, measurement models for each of the variables were explored. While confirming the measurement of a particular variable, each of these models provided factor scores to be used in generating composite factor scores from items. Using theory described by Holmes-Smith and Rowe (1994), these congeneric measurement models maximized the reliability of composite and latent variables. This was achieved by computing scale scores as linear combinations of items with factor scores as item coefficients. According to Holmes-Smith and Rowe, the composite score reliability (e.g. Cronbach alpha) is maximized if the weights on each item (i.e. coefficients) are corresponding factor scores rather than unity. Second, computed composite variables were used subsequently in structural equation modelling that examined relationships among latent variables. Munck (1979) showed that path loadings and error variances for observed variables can be fixed in structural equation modelling and that, provided correlation matrices are analysed, they are related to reliability (r) by the formulae

These formulae allow for paths from observed composite variables to latent variables and error variances of observed composite variables to be fixed. The advantage of this approach is that the number of parameters to be estimated by LISREL is sharply reduced with consequent improvement in model robustness.
Of the many indices available to report model fit, model comparison and model parsimony in structural equation modelling, three indices are reported in the present article: the Root Mean Square Error of Approximation (RMSEA), the Tucker-Lewis Index (TLI) and the Parsimony Normed Fit Index (PNFI). Whereas the RMSEA assesses model fit, the TLI and PNFI assess model comparison and model parsimony respectively. To interpret these indices, the following rules which are generally accepted in the SEM literature as reflecting good models were adopted: RMSEA should be below .05 with perfect fit indicated by an index of zero, TLI should be above 0.90 with perfect fit indicated when TLI = 1.00, and PFNI should be above 0.50 with indices above 0.70 unlikely even in a very sound fitting model. Further discussion on indices and acceptable values is provided in Byrne (1998), Kelloway (1998) and Schumacker and Lomax (1998). While the use of ¥ö2 tests to report goodness of fit of the model to the data is acknowledged as problematic in SEM, it was used in the present study to report improvements to the overall model fit as post hoc adjustments were made.
Statistics reported in the present study included squared multiple correlation coefficients (R2) for each structural equation and a total coefficient of determination (Joreskog & Sorbom, 1989). While R2 is a measure of the strength of a linear relationship, the total coefficient of determination is the amount of variance in the set of dependent variables explained by the set of independent variables. In addition to overall fit statistics, it is important to consider the strength and statistical significance of individual parameters in the model. Each path was tested using a t-test (p <.05).