Confidence intervals, precision and confounding

Publication date: August 2018Source: New Ideas in Psychology, Volume 50Author(s): David TrafimowAbstractAlthough it is well-known that confidence intervals fail to provide the probability that the population parameter of interest is within the computed interval, there nevertheless continues to be widespread support for them. Such support is based on the argument that confidence intervals measure precision; wide intervals indicate less precision whereas narrow intervals indicate more precision. But there are three types of precision; sampling precision, precision of homogeneity, and measurement precision; and confidence intervals confound them. In addition, based on the classical measurement theory and very simple mathematics, I demonstrate that it is easy to estimate all three types of precision separately. Thus, for those who are not interested in precision, there is no reason to compute confidence intervals. And for those who are interested in precision, it is better to estimate all three types of precision separately; consequently, there again is no reason to compute confidence intervals.

Source: New Ideas in Psychology - July 10, 2018 Category: Psychiatry & Psychology Source Type: research