Development of Influential Matrix for Detection of Outliers in Presence of Masking in Linear Regression using Survey Dat
96 / 83
Keywords:
Eigenvectors; Extended Conditional Cook-statistic; Influence matrix; Outlier; Masking; Complex survey data; Linear regressionAbstract
Identification of outliers and influential points is crucial in linear regression, particularly when dealing with survey data. Earlier works by Li and
Valliant (2011, 2015) proposed diagnostic measures for detecting single influential observations in both unclustered and clustered survey data.
However, these methods face challenges when multiple influential subsets exist, primarily due to the phenomenon of masking, where the presence
of multiple outliers conceals their influence. In contrast, substantial advancements for non-survey data have been made, notably by Peña and Yohai
(1995) and Lawrance (1995), who introduced diagnostic procedures capable of detecting influential subsets under masking conditions. Peña and
Yohai (1995) developed an influence matrix defined as the matrix of uncentered covariances representing the effect of deleting each observation
on the entire dataset, with normalization to include univariate Cook’s statistics on the diagonal. Candidate outliers are identified by examining the
eigenvectors associated with the non-null eigenvalues of this matrix. Building on this concept, we have developed an influence matrix tailored for
survey data, enabling the detection of influential subsets even in the presence of masking. The proposed diagnostic is demonstrated using a real-life
survey dataset, highlighting how masked outliers can significantly affect both ordinary and survey-weighted least squares estimation.
Downloads
References
1. Atkinson, A.C. and Riani, M. (2000). Robust Diagnostic Regression
2. Analysis. New York: Springer- Verlag.
3. Atkinson, A.C. and Riani, M. (2004). The forward search and data
4. visualization. Computational Statistics, 19, 29-54.
5. Belsley, D.A., Kuh, E. and Welsch, R.E. (1980). Regression Diagnostics.
6. New York: JohnWiley & Sons.
7. Binder, D.A. (1983). On the variances of asymptotically normal
8. estimators from complex surveys. International Statistical
9. Review, 51, 279-292.
10. Cook, R.D. (1977). Detection of influential observation in linear
11. regression. Technometrics, 19, 15-18.
12. Cook, R.D. and Weisberg, S. (1982). Residuals and Influence in
13. Regression. London: Chapman & Hall Ltd.
14. Deville, J.C. and Särndal, C.E. (1992). Calibration estimators in survey
15. sampling. Journal of the American Statistical Association, 87,
16. 376-382.
17. Kumar, R., Biswas, A., Singh, D. and Ahmad, T. (2024). Detection
18. of outliers in survey–weighted linear regression. Mathematical
19. Population Studies, 31(3), 147-164.
20. Kumar, R. and Bhar, L.M. (2023). Procedure for the identification
21. of
22. multiple influential observations in block design for
23. incomplete multi-response experiments in presence of masking.
24. Communications in Statistics-Simulation and Computation, 52(5),
25. 2167-2176.
26. Lawrance, A.J. (1995). Deletion influence and masking in regression.
27. Journal of the Royal Statistical Society B, 57(1), 181-189.
28. 245
29. Pena, D. and Yohai, V.J. (1995). The detection of influential subsets in
30. linear regression by using an influence matrix. Journal of Royal
31. Statistical Society B, 57, 145-156.
32. Pfeffermann, D. and Holmes, D.J. (1985). Robustness considerations
33. Li, J. and Valliant, R. (2006). Influence analysis in linear regression
34. with sampling weights. Proceedings of the Section on Survey
35. Research Methods, American Statistical Association,3330-3337.
36. Retrieved
37. from
38. https://www.asasrms.org/Proceedings/y2006/
39. Files/JSM2006-000435.pdf
40. Li, J. and Valliant, R. (2011). Detecting Groups of Influential
41. Observations in Linear Regression Using Survey Data—Adapting
42. the Forward Search Method. Pakistan Journal of Statistics, 27(4),
43. 507-528.
44. Li, J. and Valliant, R. (2011). Linear regression influence diagnostics for
45. unclustered survey data. Journal of Official Statistics, 27, 99-119.
46. Li, J. and Valliant, R. (2015). Linear regression diagnostics in cluster
47. samples. Journal of Official Statistics, 71, 61-75.
48. Neter, J., Kutner, M.H., Nachtsheim, C.J. and Wasserman, W. (1996).
49. Applied Linear Statistical Models. McGraw Hill/Irwin Series:
50. Operations and Decision Sciences (Fourth edition).
51. in the choice of method of inference for the regression analysis
52. of survey data. Journal of the Royal Statistical Society A, 148,
53. 268-278.
54. Ren, R. and Chambers, R. (2001). Studies on outlier robust estimators
55. for finite populations. Euredit project report.
56. Särndal, C.E., Swensson, B. and Wretman, J. (1992), Model Assisted
57. Survey Sampling, Springer Verlag, New York.
58. Scholl, J.R. (1997). Matrix Analysis for Statistics. New York: John
59. Wiley.
60. Skinner, C.J., Holt, D. and Smith, T.M.F. (1989). Analysis of Complex
61. Surveys. New York: John Wiley.
62. Weisberg, S. (2005), Applied Linear Regression, New York: John Wiley