Objetivo 5. Sistema de salud integrado y eficiente
2. Metodología, proceso de consulta, enfoques y valores del PDS 2022 - 2031
Following the procedure outlined in Chapter 4.3.4 describing the use of structural equation modeling in this study, I first examined the feasibility of the parameter estimates. The parameter-level examination indicates a good model fit. No correlations above 1.00, or negative variances were found (Byrne 2001:75). The covariance matrix was also positive definite. Standard errors were also reasonable and the direction and significance of the parameters were according to the underlying theories and hypotheses in sixteen out of seventeen hypothesized parameters, also suggesting good fit of the model. These analyses should reveal potential severe violations in the model fit. Based on these analyses, the model appears to behave well. The parameter estimates are further discussed in later sections discussing the results of hypotheses testing.
The next phase of the analysis is the examination of the model as a whole. As Table 5-28 demonstrates, the overall fit of the hypothesized model is good. The Chi-square test indicates a non-significant difference between the hypothesized and observed covariance matrices (p ≥.10), thus suggesting a good fit of the model. The Normed Chi-square statistic for the hypothesized model is 1.38, well within the recommended range 1.0-2.0 (Hair et al. 1998). Values close to or above .90 on the goodness-of-fit index and non-normed fit index are desirable. The hypothesized model exceeds these limits. The comparative fit index value .970 exceeds the new strict criteria of .950 thus indicating a good fit (Hu & Bentler 1999). Also root mean square error of approximation was within the recommended limits for acceptable fit of .08 (Browne &
Cudeck 1993, MacCallum et al. 1996). Overall, the hypothesized model appears to fit well in the observed data.
Table 5-28 Goodness of fit statistics for the structural equation models
Model χ2 df p Normed χ2 GFI NNFI CFI AIC RMSEA
1. Null model 312.91 45 .000 6.95 .469 .000 .000 332.91 .257 2. Hypothesized model 29.05 21 .113 1.38 .944 .936 .970 97.05 .065 3. Partial mediation model 1
(Direct path added between complementarities and knowledge acquisition)
29.03 20 .087 1.45 .944 .924 .966 99.03 .071
4. Partial mediation model 2 (Direct path added between complementarities and endorsement)
27.86 20 .113 1.39 .946 .934 .971 97.86 .066
Normed Chi-square = Chi-square adjusted by degrees of freedom, GFI = Jöreskog and Sörbom’s goodness-of-fit index, compares predicted squared residuals with obtained residuals, not adjusted by degrees of freedom; NNFI = Non-Normed Fit Index (Tucker and Lewis’ index) compares proposed model to null model, adjusted by degrees of freedom; and CFI = compares proposed model to null model, adjusted by degrees of freedom; AIC = Akaike information criterion; RMSEA = Root Mean Square Error of Approximation.
Nested model tests (Loehlin 1987:62-67) were employed to assess the fit of the hypothesized model and to test its robustness by comparing it to other alternative models. Nested model tests are a means of internally validating a hypothesized model by comparing the Chi-squares of models that differ in the number of paths hypothesized. Nested models can be derived from each other by adding or deleting paths. A significant difference in Chi-square indicates that the more complex model provides a better fit with the data (Steiger et al. 1985: 254).
I compared models 1 through 4 in Table 5-28 by using sequential Chi-square difference tests to obtain successive fit assessments (Steiger et al. 1985). Following a series of hierarchical tests, the validity of the hypothesized model was tested by showing that it is the best fitting of the theoretically meaningful models.
The four nested models compared in the analysis are: (1) a null model, in which no relationships are posited; (2) the hypothesized model; (3) a partial mediation model in which a direct path is added to the hypothesized model between complementarities and knowledge acquisition; and (4) a partial mediation model in which a direct path is added to the hypothesized model between complementarities and endorsement. Table 5-29 summarizes the testing sequence employed.
Table 5-29 Nested model testing sequence and difference tests
More Parsimonious Model Less Parsimonious Model ∆χ2 ∆df P Preferred 1. Null model vs. 2. Hypothesized model 283.86 24 <.005 Model 2 2. Hypothesized model vs. 3. Partial mediation model 1 0.01 1 >.100 Model 2 2. Hypothesized model vs. 4. Partial mediation model 2 1.18 1 >.100 Model 2
In the testing sequence, the first comparison is the comparison between the hypothesized model and the null model. The goodness-of-fit statistics (Table 5-28) and the Chi-square difference test (Table 5-29) indicate that the hypothesized model provides a better fit than the null model.
The second comparison is a robustness test testing the strength of the mediation effect of social interaction mediating the complementarity effects to knowledge
acquisition. In this comparison, the hypothesized model was compared to the partial mediation model in which a direct path was added to the hypothesized model between complementarities and knowledge acquisition. The difference in Chi-square is not significant (second row in Table 5-29), indicating that the more parsimonious, hypothesized model provides a better fit with the data than the partial mediation model.
The third comparison is a robustness test testing the strength of the mediation effect of resource acquisition mediating the complementarity effects to endorsement. In this comparison, the hypothesized model was compared to the partial mediation model in which a direct path was added to the hypothesized model between complementarities and endorsement. The difference in Chi-square is not significant (third row in Table 5-29), indicating that the more parsimonious, hypothesized model provides a better fit with the data than the partial mediation model.
Having tested all the relevant model alternatives, I conclude that the hypothesized model (Model 2) provides the best fit and terminate the model testing. Figure 5-1 presents the diagram of the hypothesized model tested using structural equation modeling.
Complementarities
Figure 5-1 Structural equation-modeling results of the hypothesized integrated model