Discussion Paper no. 05-03


This paper uses the 1998 Kenya DHS to examine the correlation of mortality risks across siblings. Previously, the random-effect parameter in the random model has been interpreted in terms of unmeasured and unmeasurable factors, suggesting the presence of death clustering. This interpretation is problematic because the concept of unobserved heterogeneity is not the same as death clustering. This paper attempts to clarify the concept of death clustering and demonstrates that the concept needs to be closely associated, and therefore examined, with the sequence of births and deaths in a family. Earlier approaches have been insensitive to sequencing both in the clarification of the concept and in the analytical methods used to examine the presence and extent of death clustering. Using binary sequence models that also incorporate unobserved heterogeneity we show that the parameter for death clustering is conceptually distinct from the parameter usually obtained for unobserved heterogeneity.