Strengths of our study included the use of nationally representative survey data and the long duration of time considered (18 years). We chose to use simple transparent bivariate methods (considering relationships between age, survey year and birth cohort effects two at a time) in conjunction with joinpoint regression methodology which allowed us to determine overall trends and to identify inflection points where trends changed direction. In addition, each survey year contained the same demographic and lifestyle variables, which enabled adjusted analyses. Alternatively, we could have applied age, period, cohort (APC) modeling (Robertson and Boyle, 1998) but did not do so because, using that method, it is not statistically possible to separately estimate and interpret individual parameters for age, period and cohort effects. Knowledge of any two components results in knowledge of the third.
