Abstract:
|
Empirical likelihood-based methods that compute a non-parametric estimate of the underlying joint distribution constrained by model or population information based parametric equations have been popular for integrating information from various sources in statistical modeling. In recent times, the same methodology has been used for the likelihood-based modeling of complex survey data as well as in Bayesian procedures, where empirical likelihood has been used as an alternative to usual parametric likelihoods. In this talk, we discuss model-based parameter estimation from complex survey data, by incorporating available sampling weights, the information contained in the prior distribution of the parameters and additional information known about the population corresponding to the model. The distribution of the data drawn with informative complex surveys differs from that in the population. We show that using empirical likelihood one can deduce several interpretable posteriors from which the estimates of the parameter values can be found. We illustrate our procedures using several simulated and real-life examples.
|