일반화추정방정식에서 결측대체 효율성 연구

다국어 초록 (Multilingual Abstract)

Comparing with using cross-sectional data, we can estimate the parameters more efficiently using panel data such as repeated measured data (Hsiao, 1985). In many medical research, we often try to collect repeatedly measured patient data over periods in order to learn their time trend if any. However, while collecting the data over time, there can be missing values at some time points for several reasons. Their missing mechanism can be also vary from MCAR(Missing Completely at Random), MAR(Missing at Random), and NMAR(Not Missing at Random) (Little and Rubin, 2002).
In this thesis, we want to investigate and compare the efficiency of several missing value imputation methods while we fit the generalized estimating equation models for panel data. We conducted simulation studies using several scenarios and analyzed real data using KOSCO (The Korean Stroke Cohort for functioning and rehabilitation) data.
We conducted the simulation study using simulation data for several scenarios and applied the idea to real data using KOSCO (The Korean Stroke Cohort for functioning and rehabilitation) data.
The results show that multiple imputation works best in terms of bias whether the missing mechanism is MCAR, MAR or NMAR.

목차 (Table of Contents)

• 표 차례 ⅲ
• 그림 차례 ⅴ
• 제 1장 서론 1
• 제 2장 연구방법 3
• 2.1 일반화추정방정식 3
• 2.1.1 일반화추정방정식 소개 3
• 2.1.2 독립추정방정식 6
• 2.1.3 일반화추정방정식 7
• 2.2 결측 자료 10
• 2.2.1 결측 구조 10
• 2.2.2 결측 패턴 12
• 2.2.3 결측 처리 방법론 13
• 제 3장 연구자료 20
• 3.1 자료 소개 20
• 3.2 결측 패턴 23
• 3.3 실제자료 분석 27
• 3.3.1 연속형 자료 28
• 3.3.2 순서형 자료 30
• 제 4장 모의실험 33
• 4.1 모의실험 설계 33
• 4.2 모의실험 결과 39
• 4.2.1 연속형 자료 39
• 4.2.2 순서형 자료 53
• 제 5장 결론 66
• 참고문헌 68
• 부록1 KOSCO 연구 중 본 논문 분석에 사용된 자료 72
• 부록2 결측 대체 방법론 별 GEE 분석을 위한 R코드 73