본문 바로가기 주메뉴 바로가기
국회도서관 홈으로 정보검색 소장정보 검색

결과 내 검색

동의어 포함

목차보기

Title page 1

Contents 1

Abstract 2

1. Introduction 3

2. Interpretation and Practicality of Ratio Edits 5

3. Limitations of Multivariate Ratio Edits Based on the Minimum Volume Ellipsoid 7

4. Multivariate Tolerance Regions 9

4.1. Multivariate Ratio Edits Based on Rectangular Central Tolerance Regions 10

4.2. Multivariate Ratio Edits Based on Mixed-Simultaneous Central Tolerance Regions 12

4.3. Tolerance Regions Based on Statistically Equivalent Blocks 13

4.4. Multivariate Ratio Edits Based on Statistical Data Depth 15

5. Numerical Study 15

5.1. Contaminated Data Generation 16

5.2. Monte Carlo Benchmarks 18

5.3. Simulation Results 19

6. Applications 26

7. Discussion 32

Acknowledgments 34

References 34

Appendix 40

A. Ellipsoidal vs Rectangular Tolerance Regions 40

B. Data Transformation 42

C. Multivariate Transformation 44

D. Jaccard Index 48

E. Contaminated Data Generation 52

F. Other Potential Univariate Ratio Edits Using Tolerance Intervals 54

G. Tolerance Intervals Algorithms 55

H. Simulations 60

I. Application 73

Tables 22

Table 1. Six-method rectangular TI results (q = 2, P = .95, trim = 0.10) over contamination levels ν ∈ {0.05, 0.10}. Volumes are scaled by the block-... 22

Table 2. Unit volumes used to scale the rectangular TI volumes in Table 1 (q = 2, P = .95, trim = 0.10), where WS=Well Separated,... 23

Table 3. RCTR simulation results for (q = 3, sides = two.sided, twp.sided, two.sided) under contamination ν = 5%. Reported are Type I error,... 24

Table 4. MSCTR simulation results for (q = 3, sides = two.sided, upper, lower) under contamination ν = 5%. Reported are Type I error,... 25

Table 5. Method-specific trimming recommendations based on Type I and Type II error patterns in Tables 3, 4, and tables in supplementary... 26

Table 6. Rectangular tolerance intervals (as (LU)) and corresponding outlier counts for five ratios, 2018. KT is a Mahalanobis-distance method... 29

Table 7. Rectangular tolerance intervals (as (LU)) and corresponding outlier counts for five ratios, 2019. KT is a Mahalanobis-distance method... 29

Table 8. Rectangular tolerance intervals (as (LU)) and corresponding outlier counts for five ratios, 2020. KT is a Mahalanobis-distance method... 30

Table 9. Rectangular tolerance intervals (as (LU)) and corresponding outlier counts for five ratios, 2021. KT is a Mahalanobis-distance method... 30

Table 10. Per-Ratio Counts by Year (2018-2021), Rectangular Central (trim=0.10, P = 0.95, α = 0.10, all two sided, B = 1000 for RCTR) 31

Table 11. Longitudinal Transitions (2018→2021), Rectangular Central (trim=0.10, P = 0.95, α = 0.10, all two sided, B = 1000 for RCTR) 32

Table 12. Normality diagnostics on the trimmed 2018 ASM data, before and after the joint Yeo-Johnson (YJ) transformation. Each cell reports... 48

Table 13. Central Rectangular TI (CRTI) simulation results for p = 2 under contamination ν = 10%. Reported are Type I error, Type II error,... 60

Table 14. Simultaneous Central Rectangular TI (SCRTI) simulation results under contamination ν = 10%, α = .05, P = 0.95. Two Sided=TS... 61

Table 15. Variation of MSCTR factor(s) under different conditions. Reported are k₁ and (when applicable) k₂ 62

Table 16. Variation of rectangular TI tolerance factor c₁ under different conditions 62

Table 17. Results for six rectangular TIs (p = 5, P = .95, trim = 0.10) over contamination levels ν ∈ {0.05, 0.10}. Volumes are scaled by the block-... 63

Table 18. Unit volumes used to scale the rectangular TI volumes in Table 17 (q = 5, P = .95, trim = 0.10), where WS = Well Separated,... 64

Table 19. Results for six rectangular TIs (p = 10, P = .95, trim = 0.10) over contamination levels ν ∈ {0.05, 0.10}. Volumes are scaled by the block-... 64

Table 20. Unit volumes used to scale the rectangular TI volumes in Table 19 (q = 10, P = .95, trim = 0.10), where WS = Well Separated,... 66

Table 21. RCTR simulation results for (p = 3, sides = two.sided, twp.sided, two.sided) under contamination ν = 5%. Reported are Type I error,... 67

Table 22. MSCTR simulation results for (p = 3, sides = two.sided, upper, lower) under contamination ν = 5%. Reported are Type I error,... 68

Table 23. (M)SCTR simulation results for (p = 3, all variables two-sided) under contamination ν = 5%. Reported are Type I error, Type II error,... 69

Table 24. SEB rectangular tolerance region simulation results (p = 3, all variables two-sided) under contamination ν = 5%. Reported are Type I error,... 70

Table 25. Spatial Depth rectangular tolerance region simulation results (p = 3, all variables two-sided) under contamination ν = 5%. Reported... 71

Table 26. Mahalanobis-depth rectangular TI results (p = 3). Type I bolded when closest to α within each contamination regime 72

Table 27. Per-Ratio Counts by Year (2018-2021) and Longitudinal Transitions (2018→2021), SEB (trim=0.10, P = 0.95, α = 0.10, all two sided,... 73

Table 28. Per-Ratio Counts by Year (2018-2021) and Longitudinal Transitions (2018→2021), Mixed-Simultaneous Central (trim=0.10, P = 0.95,... 73

Table 29. Per-Ratio Counts by Year (2018-2021) and Longitudinal Transitions (2018→2021), Spatial Depth TI Regions (trim=0.10, P = 0.95,... 73

Table 30. Per-Ratio Counts by Year (2018-2021) and Longitudinal Transitions (2018→2021), Mahalanobis Depth TI Regions (trim=0.10,... 74

Figures 4

Figure 1. An illustration of ratio edits workflow showing trimming of ratio data, construction of tolerance regions, and expert review of flagged cases 4

Figure 2. Ellipsoidal (blue circle) vs. rectangular (blue/red rectangles) reference regions. Adapted, with modifications, from Wolf and Wunderli [2015] 41

Figure 3. Rectangular outlier illustration. The dashed box shows the acceptance region x ∈ [xL, xU], y ∈ [yL, yU]. Points outside only on X... 41

Figure 4. Mahalanobis χ² Q-Q plots before and after transformation. Panel 4a shows departures from the reference line under the original scale;... 46

Figure 5. Histograms by ratio variable for the trimmed 2018 ASM data. Panel 5a exhibits strong right skew and long tails; panel 5b shows improved... 47

Figure 6/Figure 8. Joint agreement (Jaccard) - mean over 2018-2021 49

Figure 7/Figure 6. Per-ratio agreement (Jaccard) - 2018 50

Figure 8/Figure 7. Joint agreement (Jaccard) heatmaps by year (2018-2021) 51

Figure 9. Marginal densities across contamination regimes under the MCN model. Distinct color bands indicate "good" (blue), "lower" (gray),... 54

Figure 10. Pairwise scatterplots of simulated ratio variables under well-separated, moderate, and heavy contamination. Cyan points represent... 54