Risto Lehtonen

Chapter 31 - Design-based Methods of Estimation for Domains and Small Areas

Publisher Summary This chapter is devoted to the estimation for population subgroups or domains. Design-based estimation for a finite population quantity refers to an estimation approach where the randomness is introduced by the sampling design. Thus, the approach also is called randomization approach. In design-based estimation, it is emphasized that estimators should be design consistent and, preferably, essentially design unbiased at least in medium-sized samples. The estimation of ratios and quantiles, such as median, is also discussed. The availability of high-quality auxiliary information is crucial for reliable estimation for domains. Different types of auxiliary data can be used in design-based estimation for domains. The available auxiliary data can be aggregated at the population level, at the domain level, or at an intermediate level. Calibration techniques and model-assisted methods using aggregated auxiliary data offer efficient tools for design-based domain estimation. A statistician also can be in a favorable position to use unit-level auxiliary data for domain estimation. These data are incorporated in the estimation procedure by unit-level statistical models.

Systematic handling of missing data in complex study designs – experiences from the Health 2000 and 2011 Surveys

ABSTRACT We present a systematic approach to the practical and comprehensive handling of missing data motivated by our experiences of analyzing longitudinal survey data. We consider the Health 2000 and 2011 Surveys (BRIF8901) where increased non-response and non-participation from 2000 to 2011 was a major issue. The model assumptions involved in the complex sampling design, repeated measurements design, non-participation mechanisms and associations are presented graphically using methodology previously defined as a causal model with design, i.e. a functional causal model extended with the study design. This tool forces the statistician to make the study design and the missing-data mechanism explicit. Using the systematic approach, the sampling probabilities and the participation probabilities can be considered separately. This is beneficial when the performance of missing-data methods are to be compared. Using data from Health 2000 and 2011 Surveys and from national registries, it was found that multiple imputation removed almost all differences between full sample and estimated prevalences. The inverse probability weighting removed more than half and the doubly robust method 60% of the differences. These findings are encouraging since decreasing participation rates are a major problem in population surveys worldwide.

Estimation of Poverty Rate and Quintile Share Ratio for Domains and Small Areas

In the article, we consider the estimation of indicators on poverty and social exclusion for population subgroups or domains and small areas. For at-risk-of-poverty rate, we discuss indirect design-based estimators including model-assisted logistic generalized regression estimators and model calibration estimators. Logistic mixed models are used in these methods. For quintile share ratio, indirect model-based percentile-adjusted predictor methods using linear mixed models are considered. Unit-level auxiliary data are incorporated in the estimation procedures. For quintile share ratio, we present a method called frequency-calibration or n-calibration to be used in cases where aggregate level auxiliary data only are available. Design-based direct estimators that do not use auxiliary data and models are used as reference methods. Design bias and accuracy of estimators are evaluated with design-based simulation experiments using real register data maintained by Statistics Finland and semi-synthetic data generated from the EU-SILC survey.