| An extended statistical software for the estimation, prediction, and inference of a wide variety of standard econometric models is developed to analyze datasets involving a large proportion of missing information. This relies on Bayesian sampling-based approaches with data augmentation. A generalization of the Bayesian treatment of vector autoregressive models is also considered. As a direct by-product, the proposed methodology is shown to be a natural and effective way to address the problem of data interpolation from intermittent longitudinal surveys which is both conceptually simple, and computationally tractable.; We apply the interpolation methodology to bridge the gap between two national surveys on the use of private vehicles which are six years apart. This allows us to produce quarterly predictions of the three energy components (the average number of vehicles, the average distance travelled by each vehicle, and their weighted fuel consumption rate) for the intermediary period, between the surveys. Separate estimates and predictions are obtained by vehicle type: for cars and for light trucks and vans. The same technique could also be directly implemented in other contexts such as international database comparisons, population censuses, longitudinal labour force surveys, etc.; First of all, survey-based estimates are adjusted with the aim of improving their compatibility. Predicted values for the intermediate period are obtained by means of the Bayesian method of Gibbs sampling with data augmentation. In order to improve efficiency, by making use of all available empirical information, the econometric model is estimated on the basis of data from both surveys, while taking into account the middle period, between the two surveys' sampling periods, for which no data exist.; Based on explanatory variables from external sources, the aggregate simultaneous equations model is formulated to account for the relationships among the energy components. The model takes into account the dynamics involved in the three dependent variable time series. Since the data are aggregated on a quarterly basis, it also captures seasonal variations, in addition to the general trends in the series.; Several alternative specifications are compared to determine the best prediction model. The generalized vector autoregressive model is shown to yield the most precise and reliable results. Convergence of the iterative estimation process and its dependence on prior choices are assessed by means of sensitivity analyses. Complete time series produced by this empirical analysis will provide more accurate data on which the policy makers can rely.; Given that a similar survey is to be done soon, the necessity of obtaining, from such intermittent sources, complete time series estimates for the key variables from a transportation researcher's point of view becomes crucial. However, the proposed interpolation methodology is not a substitute to a well-designed data collection process, but rather a general solution to existing data gaps. |
