The effect of exposure variance and exposure measurement error on study sample size: implications for the design of epidemiologic studies.