|
Wed., Jan. 25
5:10 pm, FLO 100
Refreshments at 5:00 pm
|
| A typical area model assumes the variance V of the error term is known in order to avoid non-identifiability. The assumptions of known V almost becomes mandatory for secondary users of survey data who do not have access to any micro data for estimation of the V. In reality, however, the V are random based on sampled data. Thus, in situations when one has additional data to estimate the V, they can be used efficiently for modeling the V.
We address small area estimation problems where we have additional data to model the variance V of the error term in the area level model. Also, for robustification, we assume t-distribution of the random effects. In other words, this work considers joint modeling of small area means and variances along with t random effects. For a full Bayesian analysis, we find prior distributions for all the hyper-parameters. One of the key features in our analysis is including the degrees freedom which is usually assumed to be known before. A modified version of Jeffreys’ prior leads to posterior propriety. We show that the proposed approach performs better than other approaches in many situations through real data analysis and simulation.
|