3 Comments
User's avatar
Leon Voß's avatar

I think equation 8 is correct. Their proof in the appendix checks out. The SD correction is for simple correlations, but they're computing a correction for a beta. If you take the formula for correcting a correlation for attenuation, and you convert both sides (correction and observed) to betas using r= (sd(x)/sd(y))B, then you recover equation 8.

Expand full comment
Emil O. W. Kirkegaard's avatar

And yet it's the wrong equation, since the correction is done using sqrt(reliability).

Expand full comment
Headless Marbles's avatar

I feel like if you could just collaborate with a statistician or econometrician who deeply knows the theory of errors-in-variables models, and would just need to get up to speed on the sibling regression model, the two of you could derive a theoretical correction factor for the data generating process that you assume here.

Expand full comment