Abstract
When attempting to determine if the observed correlation between two variables can be "explained" by a third variable, many researchers resort to a significance test on the partial correlation coefficient. Several authors have pointed out how this can be misleading when the third variable is measured with error. We show how the problem can be partially overcome (to the extent that the problem is at all soluble with the given data) with knowledge or assumption about the reliability of measurement in the third variable. An alternative suggestion, that one instead consider the feasibility of a one factor fit to the observed intercorrelations, has been proposed; we show that this has little to recommend it.
Get full access to this article
View all access options for this article.