Nonparametric Statistical Matching Methods: An Application on Household Surveys in Turkey
xmlui.mirage2.itemSummaryView.MetaDataShow full item record
The use of administrative registers and sample surveys together in estimation process has a significant effect on increasing the accuracy of statistical information and reducing the response burden, time, cost and labour. However, this requires integration of data sources. Record linkage and statistical matching are two techniques improved for data integration. Record linkage is a method used when there is a perfect agreement between indicators. If there is no such variable in the data set but there are some common variables and samples of the surveys refer to same target population then statistical matching is used. Common variables between surveys are used as matching variables and fused data sets are obtained using different matching approaches such as parametric, non-parametric and mixed. Aim of this dissertation was to apply different non-parametric statistical matching methods on household based sample surveys and compare their results regarding similarity of variables’ distributions. 2014-2015 Time Use Survey of Turkey and 2014 Life Satisfaction Survey of Turkey were used in the implementation. Three non-parametric hot deck methods named as nearest neighbour distance, random and rank hot deck were used in the matching with their constrained and unconstrained versions. For each method, fused data sets were obtained using four different combinations of matching variables and four target variables. Comparisons were made both between methods, and within methods; through changes in applications for each specific method. It was found that results vary according to both matching variables and target variables. Moreover, contrary to expectations, constrained hot deck did not provide better result.
The following license files are associated with this item: