This research project develops new methods for the efficient use of auxiliary information in complex surveys, based on nonparametric regression techniques. Current practice relies on parametric regression techniques, which have good efficiency if the regression model is well specified, and which have a number of appealing operational features. The nonparametric techniques share these operational features, lose little efficiency when the parametric specification is correct, and gain efficiency when the parametric specification is incorrect. The project increases the scope of applicability of the nonparametric regression estimation approach, by considering complex survey designs, varying types of auxiliary information, and alternative smoothing techniques. Specifically, the project investigates multi-stage surveys with cluster or element-level auxiliary information; multivariate auxiliary information; and alternative smoothing techniques. Parametric and nonparametric techniques are blended using semiparametric additive models to provide a flexible tool for use in complex surveys.
Large-scale surveys are used to collect data in a wide range of fields, from studies of human populations to inventories of natural resources. Information external to the survey, such as administrative records or remote sensing, is often available. This research project makes it possible to incorporate auxiliary information easily and effectively into survey estimates, by using nonparametric regression methods. Nonparametric regression, sometimes referred to as smoothing, is widely used in other areas of statistics, but its use in survey estimation has been limited so far. The investigators show that incorporating auxiliary information into survey estimation through nonparametric regression can improve the precision of the surveys, often at reduced costs.
