| Population based mortality and morbidity data have been described by various authors in terms of additive age, period, and cohort effects. These three variables are linearly related and, therefore, their individual effects cannot be uniquely determined. However, despite the identifiability problem, age-period-cohort analysis can be a valuable means of identifying patterns of change in long-term data.;Assuming equal age slopes across residence categories allows one to test for differences in period and cohort trend. However, the validity of these tests rests on an unverifiable assumption. In the present analyses, trends for 16 of the 36 sex-site cancer categories showed no significant differences in trend by residence. Of these, 13 showed a consistent and statistically significant urban excess in cancer risk. In seven instances, statistically significant differences in trend between residence groups could be explained on the basis of medical diagnosis and detection. Most of the remaining urban/rural differences are consistent with a trend toward homogeneity in the two regions with regard to air pollution, cigarette smoking, and alcohol consumption. However, the present study reveals a number of differences in trend by residence that are not readily explained by accepted 'urban risk factors'. Urban and non-urban areas of Connecticut must still differ with respect to some other factors that contribute to the urban excess in cancer incidence.;The present study addresses some difficulties that arise in the application of age-period-cohort analysis to vital rates, focusing in particular on the comparison of time trends in two regions or groups. To this end, Connecticut cancer incidence data for the years 1940-79 is analyzed for 22 anatomic sites by urban and non-urban residence. Estimable functions of the parameters for the two residence groups are delineated and the type of comparisons that are possible between groups are discussed. The issue of power to detect period and cohort effects is examined. |
