Abstract

T
ABSTRACT
he aim of this paper is to provide detailed exploration of statistical models that provide estimates for planned and unplanned reporting domains during survey design. It is a global norm that National Statistical Offices depends on surveys to respond to the needs of stakeholders, however survey cost remains a challenge during survey design, estimation, and reporting stages. The paper reviewed literature relating to statistical approaches that may be used to provide estimates at lower (unplanned or planned) reporting domains using statistical models, data fragmentation and other methods that relates to big data. Several techniques such as Small Area Estimation0, have been outlined in various literature, however most national statistical offices (NSOs) do not rely on modelled data to respond to user-requests. The base of discussion for this paper focuses on models such as area level models (Fay-Herriot model, model with Correlated Sampling Errors, Time Series and Cross-Sectional Models and Spatial Models), unit level models (Multivariate Nested Error Regression Model1, Random Error Variance Linear Model2, Two level Model3, and Logistic Regression Models4). This exploration is paramount to any NSO to evaluate the most suitable models that can be used for unplanned reporting domains. The paper will provide a summary of the recommendations of feasible models that can be utilized efficiently in the production environment.

Key words: Small area estimation0, Fay-Herriot model, Correlated Sampling Errors, Time series and Cross-sectional models, Spatial Models, Unit Level Models, Random Error Variance Linear Model2, Two Level Model3, Logistic Regression Models4.