This note focuses on the following scenario common in real life : Data are collected on K subjects for a length of time on exposure to a binary risk factor and a binary disease outcome. The problem considered is the estimation of population attributable risk independent of time, which quantifies the marginal impact of the risk factor on the disease. At each of T time points, the population attributable risk is formulated as a function of risk prevalence rate and the logit model parameters relating the risk and the disease. Quasilikelihood methods for longitudinal data are applied to estimate these model parameters. Finally, applying the quasilikelihood for a second time, T estimates of the population attributable risk are combined. The methodology is illustrated with a real life data set.
