Quantitative analysis of variability and uncertainty with known measurement error: methodology and case study

Area of Science:

• Environmental health
• Biostatistics
• Epidemiology

Background:

• Measurement error in exposure and risk factor data can distort variability and uncertainty analyses.
• Observed data distributions differ from error-free distributions, impacting study inferences.

Purpose of the Study:

• To demonstrate a methodology for improving characterization of variability and uncertainty with known measurement errors.
• To present a practical method for constructing an error-free data set and address uncertainty from measurement error.

Main Methods:

• Utilized an observed data set with known measurement error and a measurement-error model.
• Developed a numerical method using bootstrap pairs and two-dimensional Monte Carlo simulation.
• Constructed an error-free data set to compare with observed data.

Main Results:

• Substantial differences in variability distributions and uncertainty probability bands occur between observed and error-free data when measurement error is significant.
• Failure to separate random sampling error and measurement error leads to biased estimates.
• Total uncertainty in the mean can be accurately quantified even without separating these error types.

Conclusions:

• Accurate quantification of variability and uncertainty requires accounting for measurement error.
• The proposed methodology improves the reliability of statistical inferences in the presence of known measurement errors.
• Separating error sources is not always necessary for accurate total uncertainty in the mean.