Correcting for attenuation due to measurement error

TL;DR

This paper introduces a frequentist method for accurately quantifying uncertainty in correcting correlations for measurement error, especially effective with small sample sizes, and recommends using confidence curves over intervals.

Contribution

It presents a new conservative frequentist approach with improved coverage properties and provides an R package for implementation and visualization.

Findings

Better coverage properties than existing methods for small samples

Use of confidence curves is recommended over confidence intervals

Provides an R package 'attenuation' for practical application

Abstract

I present a frequentist method for quantifying uncertainty when correcting correlations for attenuation due to measurement error. The method is conservative but has far better coverage properties than the methods currently used when sample sizes are small. I recommend the use of confidence curves in favor of confidence intervals when this method is used. I introduce the R package "attenuation" which can be used to calculate and visualize the methods described in this paper.