Decision-Making Processes in the Workplace

Given the intercorrelations of job demands, job resources and the two performance dimensions, standardized composite scores were computed prior to hypotheses and model testing (see Model 1 in Figure 1). All variables including the moderation terms have been patterned as latent factors with a single indicator. All latent factors were adjusted for random measurement error by establishing the random error variance of each construct corresponding to the product of its variance and the quantity minus its original internal consistency. Variables that considered moderator effects were constrained in accordance with Cortina et al., and standardized in order to estimate the reliability of the interaction terms. Such procedure is based on the original reliability of both variables used to form a product term and the correlation amongst the two latent variables as value for the path from the latent interaction factor to its indicator. As for all model variables, the error variance of the indicator of the latent interaction factor was set equal to the product of its variance minus its reliability. Finally, for DMCy, DEM, job demands, job resources, exhaustion, and two performance dimensions, the path from the latent variables to their corresponding observed variable was equal to the square root of reliability of the observed score. In testing the hypothetical Model 1 with all the interactions considered and performed with the maximum likelihood estimation method, fit indices suggested an acceptable model [x²(36.31, df 25, p > 0.06); GFI = 0.97; RMSEA = 0.047; CFI = 0.94] (Table 2).

TABLE 2. Goodness-of-Fit Indices (Maximum-Likelihood Estimates) for the Structural Equation Models proposed.

Consistent with our hypotheses, most of the main and moderation effects have been found significant and in the expected direction. Hypotheses 3 and 4 have been just partially confirmed: the relationship between exhaustion and extra-role performance (H3a) has been found not significant in the model, together with the expected interaction between DMCy and exhaustion toward in-role performance (H4b). Therefore, a second model (Model 2) was tested without the moderating effect of exhaustion in the DMCy in-role performance relationship. The elimination of the interaction path resulted in an increment to an acceptable CFI value and in an acceptable small increment of RMSEA value [x²(27.41, df 17, p > 0.05); GFI = 0.98; RMSEA = 0.054; CFI = 0.95].

Model 2 showed the same significant relations compared to model 1 in terms of main effects. Decision Making Competency (β = 0.45, p < 0.01), job demands (β = 0.67, p < 0.01) and exhaustion (β = -0.50, p < 0.01) were significantly related to in-role performance confirming H2a and H4a. Confirming hypothesis H1a, DEM (β = 0.76, p < 0.01) and job resources (β = 0.43, p < 0.01) were significantly related to extra-role performance. In addition, the structural equation modeling (SEM) confirmed three supposed interactions out of four involving different performance dimensions, as stated in hypotheses H1b (DEM × job resources → extra-role performance: β = 0.25, p < 0.05), H2b (DMCy × job demands → in-role performance: β = -0.58, p < 0.01) and H3b (DEM × exhaustion → extra-role performance: β = -0.46, p < 0.01). All the resulting relationships of Model 2 are graphically displayed in Figure 2.

FIGURE 2. Tested Model 2 with interactions between decision competences and performance dimensions (^∗p < 0.05; ^∗∗p < 0.01).