Statistical Analysis of Heaped Duration Data.pdfVIP

Statistical Analysis of Heap ed Duration Data By K Petoussis 1 RD Gill2 C Zeelenb erg3 Decemb er Abstract This pap er shows how heaping of duration data eg caused by rounding due to memory eects can b e analyzed If the data are heap ed Coxs partial likeliho o d approach which is often used in sur vival analysis is no longer appropriate We show how this problem can b e overcome by considering the problem as a missing data problem A variant of Coxs Prop ortional Hazard Mo del is constructed that takes heaping into account and is estimated by maximum likeliho o d using the EM algorithm with many nuisance parameters simultaneously for all parameters Ingredients of our metho d are application of the EM algorithm Cox regression and nonparametric maximum likeliho o d calculation with predicted data in each M step An example from practice where jackknife is used to estimate the variances illustrates the p ower of the new metho dology Keywords and phrases heaping duration data survival analysis PHM pr ole likelihood EM Intro duction Heaping o ccurs in many kinds of retrosp ectively obtained duration data For example heaping may o ccur in unemployment data obtained by p erio dic lab our force surveys LFS see for a discussion on the Italian LFS Anthrop ometric data on childrens age from Tanzania suer from another kind of heaping see Heaping is imp ortant in statistical analysis for 1 Postal address Department of Psychology Vrije Universiteit Amsterdam de Bo ele laan c HV Amsterdam The Netherlands email Kos Petoussispsyvunl 2 Postal address Mathematical Institute University of Utrecht PO Box TA Utrecht The Netherlands email gill