CMStatistics 2022: Start Registration
View Submission - CMStatistics
B1648
Title: Effectiveness's comparison of longitudinal imputation methods for wave nonresponse applied to LFS data, ICBS Authors:  Fatina Awad - Central Bureau of Statistics (Israel) [presenting]
Louiza Burk - Central Bureau of Statistics (Israel)
Tzahi Makovky - Central Bureau of Statistics (Israel)
Abstract: In a longitudinal panel (LP) survey, wave nonresponse occurs when responses are obtained for some but not all waves. The Labor Force Survey (LFS), ICBS, is an LP survey in which wave nonresponse manifests in two aspects: person wave nonresponse and household wave nonresponse, and both are handled by Nonresponse Weighting Adjustment in a Cross-Sectional approach. Longitudinal imputation methods use information obtained from previous waves in the missing wave data's imputation process in order to reduce estimation bias. Three longitudinal single imputation methods: Randomized Hot-Deck, K-Nearest Neighbors, and Classification and Regression Trees, were implemented in a multivariate approach to the LFS data through a simulations process. 500 random samples were drawn from the original non-missing data, and the original values of the target variables were removed. Thus, the values of the target variable were imputed according to each of the above methods to examine each method's performance. The effectiveness of these methods was tested by comparing the Precision metric, and measures for estimation quality, such as Root Mean Square Error, Mean Absolute Error and the fit between the distribution of the imputed values versus the original values.