Differential recall as a source of bias in epidemiologic research