ABSTRACT

In a study of environmental contamination, suppose Oj cancer cases are diagnosed in the j th of 426 census tracts during some specific period of time. (The lowercase letter o distinguishes this letter from the number 0.) To deal with one or more nuisance variates, it seems natural to divide each Oj by its counterpart number of expected cases, ej, where each 6j is cal­ culated by the indirect age-adjustment methods outlined in Section 11.5. Traditionally the logarithm of ratios like this are calculated. However, in many realistic applications, some ratio numerator, for example the compo­ nent Oj of log(oj/ej), assumes the value zero. This often happens when no cases of a disease occur during the period being studied.