How Measurement Error Attenuates a Correlation Toward Zero
The one direction of bias that runs the other way, so a student does not leave F thinking every reported number is inflated. Noise in either measured quantity shrinks the observed correlation, so an underlying relationship is generally stronger than the one you see. Directly relevant here because the authors list formant estimation error among their own limitations, which puts a floor under how large the formant heritabilities — 14% for F2 averaged across vowels, 9% for F1 of [a] — could have come out.
Questions this Concept answers
- Why does noise in a measured quantity shrink a correlation rather than manufacture one?
Correction for Attenuation in Classical Test Theory
The correction for attenuation, derived from classical test theory (Spearman, 1904), estimates the correlation between the true-score (error-free) variables underlying two measured variables by divid…