This paper proposes methods for the analysis of discrete-time survival data in which hazards are observed for pairs or larger cluster of observations. Discrete-time hazard models are commonly used in social research, either as approximations to continuous time models or for the representation of processes in which "time" is intrinsically discrete. As for all hazard models, unmeasured heterogeneity may bias estimates of the effects of determinants of discrete hazards, even when measured determinants of the hazards are well specified. Unmeasured heterogeneity is usually difficult to identify in univariate discrete-time models unless we make strong assumptions about the functional form of models. When data are paired--as, for example, in data on siblings or couples--then unmeasured determinants of the bivariate hazard can be controlled without strong assumptions about the functional form of the model or the shape of the hazard. This paper describes models for the analysis of discrete-time bivariate survival data that generalize the widely used univariate logistic model for discrete survival data to the bivariate case. These models provide a flexible way of controlling for unmeasured heterogeneity at the pair-level. In these models pair-level unobserved variables that affect the hazard of each member of the pair are represented as latent discrete variables. The models are estimated using a partially observed contingency table approach, which enables the analyst to convert an observed joint distribution of the survival times of each member of the pair into a partially observed joint distribution of the transitions made by each member. These models are illustrated using data on the school transitions of siblings.

>