A natural unselected community population measured forage, blood pressure and twelve blood and serum marker systems was subdivided into those who remained in the community, those who emigrated and those who died within four years after an examination during which the data were collected. Sampling designs which, to reduce variability due to measurement error and observer bias, employ means of multiple occasion blood pressure measurements introduce time lag, hence sample attrition, into longitudinal surveys and can result in age, blood pressure and marker phenotype distributions which are non-random samples of the population of inference. The non-random attrition of phenotypes results from the emigration of individuals in nuclear family units—members of which are genetically non-independent. Those emigrating and dying had higher initial systolic blood pressures (BP) than non-migrants, hence were not representative samples of the population of inference. It is concluded that mean individual BP scores acquired from longitudinal surveys can provide a biased estimate of the distribution of BP of the unselected community population.
Smith, David G. and Sing, Charles F.
"Sampling Biases in Longitudinal Genetic-Epidemiologic Surveys,"
3, Article 10.
Available at: https://digitalcommons.wayne.edu/humbiol/vol48/iss3/10