Bias in Select Mortality Investigations

The bias inherent in select mortality investigations where data are grouped by coincident or non-coincident rate intervals is analysed and compared. The key to the bias is shown to be the particular patterns of non-uniform ‘exposure frequency’ inherent in each method. It is shown that the bias using the non-coincident rate intervals is sensitive to a non-uniform distribution of new entrants by age, while the use of coincident rate intervals is sensitive to making appropriate assumptions regarding the distribution of policy anniversaries over life-time, p(s). Under both methods a significant proportion of the bias is retained after graduation. It is shown that non-coincident rate intervals may be preferred where p(s) is unknown, as in the CMI Bureau's investigations, while coincident rate intervals would be preferred where p(s) is known and taken into account in deriving the age for the estimated rates.