Response bias and risk ratios in epidemiologic studies.