여과 질문을 포함하는 설문조사의 항목무응답 대체방법 적용사례: 국민노후보장패널 자료를 이용하여

Abstract

Objectives: For the analysis of survey data with filter questions, item-nonresponse can lead to se- rious bias. We consider appropriate imputation methods that can deal with item-nonresponses in fil- ter questions and evaluate the performance of the methods using real data analysis and simulation study. Methods: In this paper, we considered two imputation methods, sequential regression impu- tation and sequential hot deck imputation, for the analysis of non-response survey data with filter questions. We reviewed these imputation methods and compare them with complete case analysis, available case analysis and mean imputation using real data set from Korean Retirement and Income Panel Study. Results: Simulation results showed that two methods performed in similar way when the distributions of additional variables were relatively symmetric. When the distributions were very skewed, sequential hot deck imputation yielded robust results. Conclusion: For the analysis of survey data with filter variables, item-nonresponse can lead to serious bias. Both suggested imputa- tion methods seems to reduce bias. However, the sequential hot deck imputation method is more robust than the sequential regression imputation method. In other words, the sequential regression imputation method based on parametric model assumption is sensitive to model misspecification.