Systematic patterns of zero exposures in event-history analysis

Users of intensity-regression techniques in event-history analysis based on occurrence and Exposure matrices frequently encounter zero-valued exposure s where there is no observation for selected combinations of factor levels, either because they are logically impossible or because the observational design leaves them out. For example, higher educational levels cannot be at tained at very young ages, and observations collected over a given calendar period cannot contain data for older birth cohorts at younger ages or data for younger cohorts at older ages. Unless the exposure matrix is too badly full of zeros, this does not necessarily cause problems for the estimation of regression parameters, but interpretation of results,nay have to be car ried out with extra care. In this paper we discuss why this is SO and show how deft handling of systematic pat tents of zero exposures can facilitate analysis.