Answers are grounded in this paper and cite their sources.
The correct individual single score for a within-S series of errors e about the target (for k trials) is E= √∑e(2) /k= √V(2) +C(2); the commonly used absolute error AE under-represents or deletes the variable error component V. In correlational analysis, the constant error score should be C absolute and V should be used unsquared in order to avoid curvilinearity; in general, r patterns across Ss are not predictable from within-Ss relations. While V and C are necessarily independent within Ss, they usually exhibit substantial correlation across Ss; evaluation of the role of each is sometimes important. Linearity of regression is demanded; it, rather than non-skewness, is shown to be the important assumption in using r. If relations involving algebraic C are of interest, the correlation index may be required because of U-shaped regression. Several common statistical misinterpretations are discussed.
No takes yet. Share an insight, caveat, or question.
Franklin M. Henry (1974) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: