disattenuated correlations


New Member
Objective: a guess about the actual correlations between two variables based on correlations between somewhat noisy variable estimates. The variable estimates result from coding by raters of open-ended answers. By averaging the ratings by my two raters I hope to reduce the noise and increase the reliability of the variable estimates. (In my view, the gradations are many enough to approximate an interval scale.)

Question: can I utilize these reliability estimates (of the average ratings) to calculate disattenuated correlations? That is, instead of looking just at the raw correlations between these average ratings, can I calculate the correlation as it would be if there were no measurement error from raters? That is, by dividing the raw correlation with the square root of the product of the reliability estimates of the variables in the correlation?

Many thanks,