Results

Statistical analysis was carried out using the SPSS software. The SPSS Amos structural equation modeling software was used to create the Structural Equation Models (SEMs).

The data was first checked for outliers using box-plot analysis. The only outliers identified were related to the years of employment, but these seem to be consistent to what is expected in practice in Serbia, so no observations needed to be removed from the dataset.


Exploratory Factor Analysis

Although research dimensions were empirically validated and confirmed in several prior studies, to the best of our knowledge, the empirical confirmation of the research instrument (i.e., questionnaire) and its constituents in the case of Serbia and South-Eastern Europe is quite scarce. Furthermore, the conditions in which previous studies were conducted could vary between research populations. Also, such differences could affect the structure of the research concepts. Thus, exploratory factor analysis (EFA) was conducted in order to empirically validate the structure of research dimensions and to test the research instrument, within the context of the research population of South-Eastern Europe and Serbia.

Using the maximum likelihood method we identified four factors, which account for 67% of the variance present in the data. The scree plot of the results of the analysis is shown in Figure 3. As the figure shows, we retained the factors above the inflection point.


Figure 3. Scree plot of the EFA results.

The communalities for the variables loading into the factors are shown in Table 3 and the questions corresponding to our variables are listed in Table 4. Initial communalities are estimates of the proportion of variance in each variable accounted for by all components (factors) identified, while the extraction communalities refer to the part of the variance explained by the four factors extracted. The model explains more of the variance than the initial factors, for all but the last variable.


Table 3. Communalities.



Table 4. Questions that build our constructs.

More detailed results of the EFA for the four factors are shown in Table 5. The unique loadings of specific items measured with the different questions in the questionnaire on the factors identified are shown in the pattern matrix (Table 6). As the table shows, each factor is loaded into by items that were designed to measure a specific construct and there are no cross-loadings. The first factor corresponds to job characteristics, second to job satisfaction, third to job involvement and the final to organizational commitment. The correlation between the factors is relatively low and shown in Table 7.


Table 5. Total variance explained by the dominant factors.



Table 6. Pattern matrix for the factors identified.



Table 7. Factor correlation matrix.


Confirmatory Factor Analysis

In the next part of our analysis we used Structural Equation Modeling to validate and improve a part of the model proposed by Locke and Latham that focuses on work characteristics, job satisfaction, organizational commitment and job involvement.

Although the EFA suggests the existence of four, not five, dominant factors in the model, diverging from the model proposed by Locke and Latham, in our initial experiments we used their original model, shown in Figure 4A, taking into account also organizational policies and procedures.


Figure 4. The evolution of our model (the path coefficients are standardized): (A) the initial model based on Locke and Latham, (B) no partial mediation, and (C) partial mediation introduced.

In this (default) model, the only independent variables are the job characteristics. The standardized regression coefficients shown in Figure 4A (we show standardized coefficients throughout Figure 4) indicate that the relationship between the satisfaction and organizational commitment seems to be stronger (standard coefficient value of 0.54) than the one between satisfaction and involvement (standard coefficient value of 0.37). The effect of job characteristics and policies and procedures on employee satisfaction seems to be balanced (standard coefficient values of 0.31 and 0.30, respectively).

The default model does not fit our data well. The Comparative Fit Index (CFI) for this model is 0.759, the Tucker-Lewis Index (TLI) is 0.598, while the Root Mean Square Error of Approximation (RMSEA) is 0.192.

A more detailed analysis of the model revealed that it could indeed (as the EFA suggests) be improved by eliminating the organizational policies and procedures variable, as it has a high residual covariance with job involvement (−3.071) and organizational commitment (−4.934).

We therefore propose to eliminate the "Organizational policies and procedures" variable from the model. Dropping the variable resulted in an improved model shown in Figure 4B. The improved model fits the data better, but the fit is still not good ( RMSEA = 0.125, CFI = 0.915 and TLI = 0.830).

We then hypothesized that job involvement influences organizational commitment, yielding the final model tested in this study (Figure 4C). This model turned out to be the one that fits our data very well (RMSEA = 0.000, CFI = 1 and TLI = 1.015).