Time-consistent Estimates for the Household Finance and Consumption Survey – Small Area Estimation in the Panel Context (TESAP)
Final Report Abstract
In order to target resources and interventions where they are most needed, policy makers need reliable socio-demographic indicators for regional or demographics subgroups. Because sample sizes at disaggregated levels can become small or unavailable, estimates based on survey data alone can no longer be considered reliable. Small Area Estimation (SAE) methods aim to overcome this problem and achieve higher precision. SAE methods enrich information from survey data with data from additional sources, such as administrative or census data and by using area-specific structures. Despite the problem of small sample sizes at the disaggregated level, surveys often suffer from high non-response. A possible solution to item non-response is multiple imputation, where missing values are replaced by multiple plausible values. The missing values and their substitutions introduce uncertainty into the estimation. The main purpose of this renewal work was to extend the results of the project QUESSAMI (2016-2018), which combined multiple imputation and SAE. The project QUESSAMI provided one solution for estimating mean values. In the project TESAP, we provided an adjusted and extended solution for the estimation of mean values and additionally to estimate ratios, which account for the uncertainty introduced by missing values in the SAE estimator. Because financial assets and wealth are sensitive topics, surveys of this type of data suffer particularly from item non-response. The focus was on estimating private wealth at the regional level for Germany and on estimating financial assets at the national level for EU countries. The applications use data from the 2010 Household Finance and Consumption Survey (HFCS) and data from the 2017 HFCS for European Union countries. The approach is general and can be extended to other indicators such as the Gini coefficient or applied to other survey, such as the German Sociol-Economic Panel (SOEP). To further obtain the best possible SAE estimator in terms of accuracy and precision, it is important to find the optimal model for the relationship between the target variable and the auxiliary data. One way to look at optimality is to find the best transformation of the target variable to fully satisfy the model assumptions or to account for nonlinearity. One focus of this project was to estimate Gini coefficients for regionally disaggregated data for Germany using an appropriate transformation. Another perspective is to identify the most important covariates and their relationship to each other and to the target variable. Covariates are additional data sources or time-dependent variables from previous panel waves. Approaches for model selection and optimal transformations are presented to satisfy normality assumptions and to account for interactions and nonlinear relationships.
Publications
-
Area-level small area estimation with random forests. Working Paper.
Harmening, S., Runge, M. & Schmid, T.
-
The Fay–Herriot model for multiply imputed data with an application to regional wealth estimation in Germany. Journal of Applied Statistics, 49(13), 3278-3299.
Kreutzmann, Ann-Kristin; Marek, Philipp; Runge, Marina; Salvati, Nicola & Schmid, Timo
-
Variable selection using conditional AIC for linear mixed models with data-driven transformations. Statistics and Computing, 33(1).
Lee, Yeonjoo; Rojas-Perilla, Natalia; Runge, Marina & Schmid, Timo
-
Estimating Intra-Regional Inequality with an Application to German Spatial Planning Regions. Journal of Official Statistics, 39(2), 203-228.
Runge, Marina
-
Small Area with Multiply Imputed Survey Data. Journal of Official Statistics, 39(4), 507-533.
Runge, Marina & Schmid, Timo
