Aim: Maori and Pacific deaths are under-counted in mortality data relative to census data. This 'numerator-denominator' bias means that routinely calculated mortality rates by ethnicity are incorrect. We used New Zealand Census-Mortality Study data to quantify the bias from 1981 to 1999.
Methods: The 1981, 1986, 1991 and 1996 Censuses were each anonymously and probabilistically linked to three years of subsequent mortality data, allowing a comparison of ethnicity recording.
Results: Compared with death registrations, 16% more 0-74 year old decedents during 1981-1984 had self-identified as '1/2 or more Maori' on the 1981 Census, and 32% more during both 1986-1989 and 1991-1994 had self-identified as 'sole Maori' on the 1986 and 1991 Censuses. From September 1995, mortality data have allowed multiple ethnicity to be recorded. During 1996-1999, 7% more decedents identified Maori as one of their ethnic groups on the 1996 Census compared with mortality data. For Pacific decedents, 55%, 76% and 68% more self-identified as 'sole Pacific' on census data compared with data recorded on death registrations for 1981-1984, 1986-1989 and 1991-1994 respectively, but there was no difference for 1996-1999. The bias for Maori (but not for Pacific) was greater among the young and those living in central and southern regions of New Zealand.
Conclusions: The 1995 change to ethnicity recording on mortality data has improved the robustness of ethnicity data collection. These adjustment factors for 1981-1999 allow for more accurate calculations of ethnic-specific mortality rates over the last 20 years.