Organisational citizenship behaviour might increase the efficiency of an organisation by enhancing co-worker or managerial productivity. For example, experienced employees who voluntarily help new co-workers learn to become productive employees faster, thus enhance the efficiency of the work group or unit (Alizadeh et al., 2012; Mackenzie, Podsakoff, & Fetter, 1991; Podsakoff & Mackenzie, 1994). By assisting other employees, this enables the manager to spend more time on productive tasks such as strategic planning. Individuals who engage in OCB endorse and promote the organisation to outsiders and contribute to its good reputation, attracting good candidates in the process. These individuals also help the organisation adapt to the ever changing environment, for example, when employees, who are in close contact with the marketplace, volunteer information about changes in the environment and make suggestions about how to respond, they help an organisation to adapt. This also leads to the improvement of network ties, information transfer, organisational learning and the execution of organisational activities. In short, as Podsakoff et al. (2000, pp. 543–546) documented OCBs can gain an organisation a competitive advantage by:

The promotion and creation of workplace environments that foster OCBs is, therefore, vital and to ascertain whether or not organisations are correctly assessing the exhibition of OCBs, a reliable and valid instrument is, therefore, required. The major contribution of the current study lies in ascertaining the psychometric properties of one of the widely used OCB measures.

Various organisational citizenship behaviour dimensions exist depending on the type of conceptualisation chosen. The different conceptualisation themes are depicted in Table 1. The theoretical conceptualisation underlying the Podsakoff et al. (1990) measure is based on the Organ (1988) definition hinged on ‘altruism, conscientiousness, sportsmanship, civic virtue, and courtesy’. This is arguably the most widely used conceptualisation of OCB in the literature. The Organ (1988) dimensions are defined as follows:

The objectives set out for this study were achieved through the use of structural equation modelling (SEM). A quantitative survey design was used to achieve these research objectives.

Structural Equation Modelling (SEM) is a covariance technique that explain relationships between manifest variables and their underlying latent variables (measurement models) as well as hypothesised relationships in structural models (Diamantopoulos & Siguaw, 2000).

LISREL 8.80 (Jöreskog & Sörbom, 2006) was used to perform first- and second-order confirmatory factor analysis on the OCBS to determine the fit of the model. The measurement model was considered as an exogenous model for the aims of confirmatory factor analysis. Robust maximum likelihood (RML) was used as the method of parameter estimation (Jöreskog & Sörbom, 1996).

Evaluation of fit of the first- and second-order models was based on the root mean square error of approximation (RMSEA); Root Mean Squared Residual (RMR); the Goodness-of-fit index (GFI); Normed Fit Index (NFI); Non-normed Fit Index (NNFI); Comparative Fit Index (CFI); Incremental Fit Index (IFI) and the Relative Fit Index (RFI).

The root mean square error of approximation (RMSEA) is a measure of closeness of fit and is generally regarded as one of the most informative fit indices. When assessing the RMSEA, values less than 0.05 are indicative of good fit, those between 0.05 and under 0.08 of reasonable fit, values between 0.08 and 0.10 indicate mediocre fit and those above 0.10 indicate poor fit (Diamantopoulos & Siguaw, 2000).

The root mean squared residual (RMR), is a summary measure of fitted residuals and presents the average value of the difference between the sample covariance (variance) and a fitted (model-implied) covariance (variance). The disadvantage of the RMR statistic is that the RMR varies from variable to variable. This problem is resolved by concentrating on the standardised RMR. When assessing the standardised RMR, values below 0.05 are indicative of acceptable fit (Diamantopoulos & Siguaw, 2000).

The goodness-of-fit-index (GFI) shows how closely the model comes to perfectly reproduce the observed covariance matrix. Acceptable values of the GFI should range between 0 and 1 with values greater than 0.90 being interpreted as indicating acceptable fit (Diamantopoulos & Siguaw, 2000).

The normed fit index (NFI) and the non-normed fit index (NNFI) as well as the comparative fit index (CFI) should range between 0 and 1, with values closer to 1 representing good fit (Diamantopoulos & Siguaw, 2000).

Missing values were addressed using the multiple imputation method (Jöreskog & Sörbom, 2006). In this method, missing values are replaced by values derived from averages obtained via simulation (Jöreskog & Sörbom, 2006; Rubin, 1987). The use of this technique resulted in an effective sample size of 503 cases.

The measurement and structural models were investigated by means of confirmatory factor analysis (CFA) and structural equation modelling through LISREL 8.80 (Du Toit & Du Toit, 2008; Jöreskog & Sörbom, 2006) to evaluate the construct validity of the measurement models. The present study intended to determine whether or not the original factor structure developed in the United States and Canada by Podsakoff et al. (1990) can be confirmed on a South African sample.

Jöreskog and Sörbom have established that ‘the default method of estimation when fitting measurement models to continuous data (maximum likelihood) assumes multivariate normality’ (Jöreskog & Sörbom, 1996). Both the null hypothesis of univariate and multivariate normality had to be rejected in the case of all the individual item indicator variables (p < 0.01). Thus, the use of robust maximum likelihood estimation was subsequently explored as the assumption of a multivariate normal distribution did not hold (Mels, 2003).

Item analysis using the SPSS Reliability procedure (SPSS Inc., 2014) was performed on the items of the Organisational Citizenship Behaviour Scale (OCBS). The internal consistency coefficients of the subscales of the OCBS were adequate (α ≥ 0.70) (Nunnally & Bernstein, 1994; Pallant, 2010) (See Table 2).The correlations amongst the five latent OCB dimensions were high above 0.90 with the exception of the sportsmanship subscale which was within reasonable limits, indicating a high possibility of multi-collinearity in the case of the other four subscales (Tabachnick & Fidell, 2001).

As indicated in Table 3, the RMSEA suggested that the first-order and second-order measurement models showed reasonable model fit with the obtained data, as values < 0.05 represent good fit whilst those between 0.05 and 0.08 indicate reasonable model fit (Diamantopoulos & Siguaw, 2000). The test of close fit indicates that the first-order model shows close fit (p > 0.05) whilst the second-order model does not show close fit. The standardised RMR values of 0.044 for both the first-order and second-order models reached the < 0.05 level indicative of good model fit. The GFI values for the first-order and second-order measurement models fell marginally below the 0.90 level indicative of good fit.

In terms of the goodness-of-fit of the competing one-factor model, the RMSEA indicates poor model fit as the value (0.088) (see Table 3) is above 0.08 cut-off level for reasonable model fit (Diamantopoulos & Siguaw, 2000). The test of close fit indicates that the one-factor model does not show close fit (p > 0.05). The standardised RMR values of 0.061 also indicate poor model fit whilst the relative fit measures, namely, the NFI, NNFI, CFI, IFI and RFI indices > 0.90, which represent good fit (Hair, Anderson, Black, Babin & Black, 2010; Kelloway, 1998). The GFI value of 0.78 is below the recommended cut-off value of 0.90 indicative of good fit (Diamantopoulos & Siguaw, 2000). The five-factor model appears to portray better model fit indices compared to the one-factor model.

The measurement models achieved NFI, NNFI, CFI, IFI and RFI indices > 0.90, which represents good fit (Hair et al., 2010; Kelloway, 1998). These relative fit indices generally indicate good model fit.

The completely standardised factor loadings are shown in Table 4. The factor loadings of the items are generally significant (> 0.50) with the exception of one item (Item 16) for the sportsmanship subscale with a loading of 0.40. Figure 1 indicates the factors loading on each of the OCB sub-dimensions.

The unstandardised Gamma matrix depicts the significance of the estimated path coefficients which express the strength of the influence of the exogenous latent variable (OCB) on the endogenous latent variables (the manifest variables of OCB). The parameters are significant (p < 0.05) if t-values are ≥│1.96│ (Diamantopoulos & Siguaw, 2000). The t-values show that the five dimensions are significant indicators of the organisational citizenship behaviour higher-order factor, as the t-values are greater than 1.96. The results are shown in Table 5.

The Rweb syntax compiled by Preacher and Coffman (2006) indicated a power value of one for the test of close fit which has significant implications on the rejection of incorrect models developed and tested under more or less similar conditions.

Discriminant validity was assessed using the 95% confidence intervals utilising an Excel macro developed by Scientific Software International (Mels, 2010). The use of this macro indicates that four of the OCB subscales raise some questions regarding their discriminant validity. The conscientiousness, courtesy, civic virtue and altruism subscales in the present study correlated above the 0.90 level, and their estimates in the macro (see Table 7) are close to unity as a lack of discriminant validity is indicated by the intervals that are close to one (Mels, 2010).

Future studies should determine the measurement equivalence of the OCBS across different South African gender and cultural groups. In addition, the convergent and divergent validity of the OCB should be determined in future studies, linking the construct and its sub-dimensions with other related constructs. Furthermore, there is a need to replicate the study using public and private sector employees to establish if similar results would be obtained. Four of the subscales correlate closely in this study, which suggests a multicollinearity problem. Future studies should consider collapsing the four subscales into one dimension to improve the discriminant validity amongst the subscales.

The psychometric evaluation of the OCBS in the present study indicates reasonable construct validity but limited or questionable discriminant validity. The practical contribution of the study is in the advancement of the body of knowledge on the psychometric properties of the OCBS, on a sample comprising of teachers from the Western Cape and academic and non-academic members of staff at a university in the Eastern Cape Province of South Africa. The study contributes to the advancement of the use of valid and reliable instruments as required by the Employment Equity Act (No. 55 of 1998) and the Amended Employment Equity Act of South Africa (Republic of South Africa, 1998), which require all test developers and users to consider the psychometric properties of measures before they can be used in South Africa. The OCBS demonstrated some promising evidence of reliability and construct validity and can contribute to the scientific selection and development of employees in South African educational institutions.