Sankhya: The Indian Journal of Statistics

2003, Volume 65, Pt. 1, 72--90

Rank Estimation For Longitudinal Data With Subject Attrition Due To Response Selection


Z.D. Bai, North~East~Normal~University,~China~And~National~University~Of~Singapore,

Michael Chen, Wyeth Consumer Healthcare, Madison, Usa And Gang Li, Organon Inc., West Orange, Usa

SUMMARY. This paper proposes a rank type estimation of the covariate coefficients for longitudinal data with subject attrition due to response selection. For example, to evaluate aortic aneurysm growth, the aortic aneurysm is periodically monitored and will be removed surgically once its measurement exceeds a predetermined threshold. Due to this response selection, the number of observations per subject is random because of its dependence on the measurement. As a result, the least squares estimator for the covariate coefficient is inconsistent.  This paper shows, however, that the rank estimator is consistent and asymptotically normal.

AMS (1991) subject classification. Primary 62G08; secondary 62J05, 62N02.

Key words and phrases. Rank tests, estimating equations.

Full paper (PDF)