Inference procedures based on the partial likelihood function for the Cox p
roportional hazards model have been generalised to the case in which the da
ta consist of a large number of independent small groups of correlated fail
ure time observations (Lee, Wei & Amato, 1992; Liang, Self & Chang, 1993; C
ai & Prentice, 1997). However, the Cox model may not fit the data well. A c
lass of linear transformation models, which includes the proportional hazar
ds and odds models as special cases, has been studied extensively for univa
riate event times. In this paper, statistical methods to analyse such corre
lated observations are proposed for these models. We use the data from a re
cent study of the genetic aetiology of alcoholism to illustrate the new pro
cedures for estimation, prediction and model selection.