Classroom Management Self-Efficacy and Burnout: A Multivariate Meta-analysis

Abstract

Like many in the human services professions, teachers are susceptible to the feelings of burnout due to their job demands, as well as interactions with students, colleagues, administrators, and parents. Many studies have identified teacher burnout as one of the crucial components influencing teacher attrition. It has been suggested that self-efficacy is a protective factor against burnout. By way of multivariate meta-analysis, we examined the evidence for classroom management self-efficacy (CMSE) in relation to the three dimensions of burnout: emotional exhaustion, depersonalization, and (lowered) personal accomplishment. Results from sixteen studies indicate that there is a significant relationship between classroom management self-efficacy and the three dimensions of burnout, suggesting that teachers with higher levels of CMSE are less likely to experience the feelings of burnout. Practical implications, as well recommendations for future research, are discussed.

Appendix

The framework for the multivariate meta-analysis is presented in the context of associated multi-parameters (Gasparini et al. 2012; Jackson et al. 2010; Jackson et al. 2012). The outcomes (i.e., effect sizes) represented as y _i is a p × 1 and assumed to be multivariate normally distributed with mean X _i β and k × k variance–covariance matrix (Σ _i ). For a model with no predictors, X _i is an identity matrix (p = 1) and β is k intercepts. Also, variance–covariance matrix (Σ _i ) is given by the sum of the within-study and between-studies variance–covariance matrices such as Σ _i  = S _i  + Ψ. When predictors are included in the model, X _i becomes a k × kp design matrix and β is the vector of the fixed-effect coefficients. An extension of the typically univariate test of homogeneity of effect sizes can be written as (Becker 1992; Jackson et al. 2010)

$$ {Q}_T={\displaystyle \sum_{i=1}^k\left\{{\left({\mathbf{y}}_i-{\mathbf{X}}_i\boldsymbol{\upbeta} \right)}^T{\mathbf{S}}_i^{-1}\left({\mathbf{y}}_i-{\mathbf{X}}_i\boldsymbol{\upbeta} \right)\right\}}, $$

where under the null, it follows a chi-square distribution with n − q degrees of freedom, where n = kp (the total number of observations) and q is the number of fixed-effect coefficients in β. A significant value indicates the effect sizes do not all agree and a model that incorporates between-effects variability should be adopted. In addition, a significant Q _Resididual indicates that there is still remaining variation among the effect sizes after the inclusion of a potential predictor variable in the model. The within- and between-study stages of the analysis follow.

Stage-One (Within-Study) Analysis

The Pearson product-moment correlations (r _ij ) between the four variables of interest [classroom management self-efficacy, emotional exhaustion, depersonalization, and (lowered) personal accomplishment) were extracted from each study. Then the reliability information from each study was used to correct the correlations for attenuation (Hunter and Schmidt 1990). Specifically, $E{S}_{ij}={r}_{ij}/\sqrt{r_{ii}}\sqrt{r_{jj}}$, where ES _ij is the unattenuated correlation, r _ij is the correlation extracted for the manuscript between variable i and variable j, and r _ii and r _jj are the reliabilities associated with variables i and variable j, respectively. The variance for each ES _ij was computed as V _r  = v(r)/a ², where v(r) = ((1 − r _ij ²)²)/(n − 1), and a = r _ij /ES _ij (Borenstein et al. 2009 p. 343). The asymptotic covariance between each pair of ES _ij was estimated using the asymptotic covariance between correlations (Becker 2000; Olkin and Siotani 1976, p. 238) as

$$ \begin{array}{c}\hfill \operatorname{Cov}\left({r}_{ist},{r}_{iuv}\right)=\left[0.5{\rho}_{ist}{\rho}_{iuv}\left({\rho}_{isu}^2+{\rho}_{isv}^2+{\rho}_{itu}^2+{\rho}_{itv}^2\right)+{\rho}_{isu}{\rho}_{itv}+{\rho}_{isv}{\rho}_{itu}\right.\hfill \\ {}\hfill -\left.\left({\rho}_{ist}{\rho}_{isu}{\rho}_{isv}+{\rho}_{its}{\rho}_{itu}{\rho}_{itv}+{\rho}_{ius}{\rho}_{iut}{\rho}_{itv}+{\rho}_{ivs}{\rho}_{ivt}{\rho}_{ivu}\right)\right]/{n}_i,\hfill \end{array} $$

where r _ist is the correlation between the sth and tth variables in study i, r _iuv is the correlation between the vth and uth variables, and ρ _ist and ρ _iuv are the population values. Our analyses were conducted in the unattenuated correlation metric. Although Hedges and Olkin (1985) recommended the use of Fisher’s z transformation when combining correlation coefficients, there is no complete agreement in the literature if correlations should be transformed to Fisher’s z. A detailed discussion of this issue is outside the scope of this manuscript; the interested reader is directed to Field (2001), Hafdahl (2009), Hunter and Schmidt (1990), and Hunter, Schmidt, and Jackson (1982).

Stage-Two (Between-Studies) Analysis.

Once the three effect sizes (ES _ij ) were extracted, the three variances and three asymptotic covariances were computed; the mvmeta package (Gasparrini et al. 2012) was used to perform our analyses (R Core Team, 2012). First, we estimated the overall weighted effect size for each outcome and the between-studies variance–covariance matrix (maximum likelihood estimators of between-studies variation were employed in this study). Then moderator variables were included in our analyses (e.g., study publication year, average teachers’ years of experience, percentage of females, publication type, and country of origin). When appropriate, dummy variables were created.