Entrepreneurial size, complexity and decentralization of decision-making in the use of temporary help workers in Spain

Abstract

Using data from a postal survey of Spanish firms, the present work, as its main objective, tries to evaluate the effect of entrepreneurial size, degree of decentralization in decision-making and vertical complexity on the use of temporary help workers. As complementary objectives, it also seeks to confirm the effect of entrepreneurial size on elements of the organizational structure, such as decentralization in decision-making, and on vertical complexity. To respond to the hypotheses, a LISRE model has been used. The results show that entrepreneurial size and degree of decentralization positively affect the frequency with which companies use temporary help workers. In contrast, vertical complexity negatively affects the use of this type of worker. On the other hand, it is confirmed that entrepreneurial size positively affects decentralization in decision-making and vertical complexity.

Keywords:

• temporary help workers
• centralization
• vertical complexity
• entrepreneurial size
• Spain

Notes

1 This act has since been modified by Spanish Parliamentary Act 29 (1999).

2 We should bear in mind that we are assuming a positive linear relation between organizational size and level of vertical complexity, when it is possible that the relationship is positive but that the increase in vertical complexity tapers off as size increases (Hall 1996, p. 95). This would be logical if we consider that when an organization grows in size it becomes increasingly formalized in order to coordinate and control its activity. This partially frees the supervisors from these tasks, so that they can concentrate on supervision of an increased number of staff.

3 A principal components factor analysis of the questionnaire measurement items yielded four factors with eigenvalues greater than 1.0, which accounted for 67% of the total variance. Since several factors – as opposed to one single factor – were identified and since the first factor did not account for the majority of the variance (28.84%), there does not appear to be a substantial amount of common method variance (Podsakoff and Organ 1986).

4 The PRELIS processor – for a significance level of 5% – proved the existence of significant differences both in asymmetry (p = 0.001) and kurtosis (p = 0.001). The normality condition requires a joint evaluation of the asymmetry and kurtosis level, resulting in a joint degree of the normal (χ² = 136.467 for p = 0.0001). As it did not fulfil the normality condition, neither Maximum Likelihood (ML) nor Generalized Least Squares is recommended to estimate the measurement model. Therefore, the Weighted Least Squares (WLS) procedure of the LISREL 8.30 software was deployed, which involves the computation of the polychoric matrix and the asymptotic variance matrix.

5 In particular, discriminant validity was established between each two latent variables by constraining the estimated correlation parameter between them to 1.0 and then performing a chi-square difference test on the values obtained for the constrained and unconstrained models (see Anderson and Gerbing 1988). The resulting significant differences indicate that the constructs are not perfectly correlated and that discriminant validity is achieved.

6 Among the absolute fitness measures, the chi-square value is 42.99 at a significance level of 0.014, which means that the model fits the data. Other indices must also be analysed. The Goodness of Fit Index (GFI) is another measure of goodness of fit, which explains the model variability. The GFI value can range from zero to 1, being closer to 1 when fitness is good, and closer to zero in the contrary case. In our model GFI reaches 0.96.

Values of the Root Mean Square Error of approximation (RMSEA), which should be below 0.08, is 0.073 in our model. The estimated Non-Centrality Parameter (NCP) and Expected Cross-Validation Index (ECVI) are recommended to compare different models; these are better at lower values. In this case, the values are low.

The incremental measures of fitness are above the acceptance levels suggested by empirical research. The Adjusted Goodness of Fit Index (AGFI), Comparative Fit Index (CFI) and Incremental Fit Index (IFI) should have values higher than 0.9, and reach 0.93, 0.91, and 0.91, respectively, in the proposed model.

Finally, parsimony fit measures indicate the level of fit per estimated coefficient and are very suitable for alternative model comparison. Normed chi-square (1.71) has a value within the level of good fit, which ranges from 1 to 3. Finally, values of the Parsimony Goodness of Fit Index (PGFI) and Parsimony Normed Fit Index (PNFI) can be used to compare models, with higher values being preferred.

7 There is an indirect effect (.01, p>.10) of size on THW use, via vertical complexity (.29 × − .17) and decentralization (.23 × .25); see, for instance, Bollen (1989).

8 Thus, for example, if we compare the theoretical model (Model 1) with a model that does not consider the relation between size and THW use (Model 4), we can see that the latter has a worse Expected Cross-Validation Index (ΔECVI = .01), Akaike Information Criterion (ΔAIC = 41.78), Estimated Non-Centrality Parameter (ΔNCP = 42.78) or Root Mean Square Error of approximation (ΔRMSEA = .09). Hence, the results show that size affects THW use and Model 1 is preferred to Model 4 (Δ χ² = 11.49, Δdf = 1). Likewise, we see that the theoretical model is preferred to the remaining models formulated.