Background: Socioeconomic differences in mortality in New Zealand have traditionally been measured using occupational class from mortality data (based on usual or last occupation) as the numerator, and class from census data (current occupation on census night) as the denominator. Such analyses are prone to numerator-denominator bias. Record linkage of census and mortality data in the New Zealand Census-Mortality Study (NZCMS) allows analyses of 'linked' data that will avoid numerator-denominator bias, but may be prone to other biases.
Objectives: To determine differences in the assignment of occupational class between census and mortality data; to investigate biases in the observed association of class with mortality using linked census-mortality data; and to compare the class-mortality association using unlinked versus linked census-mortality data.
Methods: Census records for males aged 25-64 years on census night 1991 were anonymously and probabilistically linked to 5,844 out of 8,145 eligible deaths occurring in the second and third years following census night.
Results: (by objective) Only 47% of linked deaths had an occupation recorded on census data, compared to 84% on mortality data - a census to mortality ratio of 0.56. Relatively fewer deaths were identified as class 4 on census data (census to mortality ratio of 0.45) compared to other classes (ratios 0.55 to 0.64). Linkage bias: A lower likelihood of 25-44 year old deaths (but not 45-64 year olds) from lower socioeconomic classes being successfully linked to a census record meant that analyses using linked census-mortality data underestimated the class-mortality association. Bias due to exclusion of economically inactive: Analyses on linked-census data (using current occupational class) considerably underestimated the estimated association of usual occupational class with mortality. The strength of the association of class with mortality according to linked census-mortality data (and adjusted for the above two biases) and unlinked data were roughly comparable.
Conclusion: Differences in the recording of occupational class on census and mortality data in New Zealand mean measuring mortality differences by class is thwart with difficulty. If one assumes that biases for any particular method of analysis are similar over time, or one carefully adjusts where possible for bias, using occupational class to monitor trends in socioeconomic mortality gradients may be valid.