An application of some bayesian diagnostics to a study of the impact of blood lead upon human blood pressure

Recently, Pirkle et al. used a sample of 40–59 year old white males from the second cycle of the U.S. National Health and Nutrition Survey (NHANES II) to estimate the association between blood pressure and blood lead concentration. They found that, on average, a 1 μg/dl increase in blood lead increased diastolic blood pressure by about 4.5 per cent and systolic blood pressure by about 6.5 per cent. We use Bayesian diagnostics to enquire into the statistical robustness of some dimensions of their results. In particular, we ask whether the results they report are sensitive to specification uncertainty and are intolerant to measurement error. We conclude the possibility to be remote that the introduction or the omission of other covariates will significantly alter the estimated influence of blood lead upon either the diastolic blood pressures or the systolic blood pressures of the adult white males in their sample. Similarly, our analysis demonstrates that other covariates are likely to affect the estimated influence of blood lead concentration only if they are quite poorly measured.