PATH ANALYSIS MODELS VERSUS LATENT VARIABLEMODELS: EXAMPLES FROM EDUCATIONAL ANDHEALTH APPROACHES

摘要

The aim of this study is to compare partially disaggregated structural models, usingitem parcels as indicators of latent variables, and path analysis models, assuming totallyaggregated items make up measurements without errors of the constructs. Data comefrom two empirical studies: one focused on education and another one focused on health,in such a way we can prove the constructs evaluated do not interfere with methodologicalresults showed. The results in both studies are consistent. On one hand, the use of itemsparcels as indicators of the constructs rather than path models increase the value of theparameters in absolute value, and residual variance is reduced. This is reflected in anincrease of the explained variance by the models. Parcel construction method does notseem to be a relevant factor. On the other hand, goodness-of-fit indexes of path modelsshow a better fit. Summarizing, latent variable models would be better fitted to reality,improving the percentage of explained variance, but losing fit due to the higher numberof parameters to be estimated.