Model-based analysis in survey: an application in analytic inference and a simulation in Small Area Estimation

This paper addresses model-based procedures in data analysis for national complex data and a simulation study in small area estimation. For National Agriculture Workers Survey data, we do variable selection and want to find one model based on Akaike information criterion (AIC). We use augmented model and weighting approach to deal with the survey weight, equal weight, and smoothed weight. Research result indicates that the survey weight plays an important role in model selection. We also show what variables are significant predictors for the number of years farm workers are employed in their current employer. For small area estimator, we conduct a simulation to compare two direct estimators and one sample mean estimator in the area-level model and one EBLUP in the unit-level model. We are interested in the mean squared error of the area-level and unit-level estimators. The research result shows that the unit-level estimator performs the best. For a larger number of sample size in each area, a regression estimator in an area-level model will be as efficient as the unit-level estimator