Holding out for a reliable change from confusion to a solution: A
comment on Maassen's “The standard error in the Jacobson and
Truax Reliable Change Index.”
It is important to preface this piece by advising the reader that the author is not writing from the point of view of a statistician, but rather that of a user of reliable change. The author was invited to comment following the publication of an original inquiry concerning Reliable Change Index (RCI) formulae (Hinton-Bayre, 2000) and after acting as a reviewer for the current Maassen paper (this issue, pp. 888–893). Having been a bystander in the development of various RCI methods, this comment serves to represent the struggle of a non-statistician to understand the relevant statistical issues and apply them to clinical decisions. When I first stumbled across the ‘classical’ RCI attributed to Jacobson and Truax (1991) (Maassen, this issue, Equation 4), I was quite excited and immediately applied the formula to my own data (Hinton-Bayre et al., 1999). Later, upon reading the Temkin et al. (1999) paper I commented on what seemed to be an inconsistency in their calculation of the error term (Hinton-Bayre, 2000). My “confusion” as Maassen suggests was derived from the fact that I noted the error term used was based on the standard deviation of the difference scores (Maassen, Expression 5*) rather than the Jacobson and Truax formula (Maassen, Expression 4). This apparent anomaly was subsequently addressed when Temkin et al. (2000) explained they had employed the error term proposed by Christensen and Mendoza (1986) (Maassen, Expression 5). My concern with the Maassen manuscript was that it initially appeared two separate values could be derived through using expressions 5 and 5* using the Temkin et al. (1999) data. This suggested there might be four (expressions 4, 5, 5*, and 6), rather than three, ways to calculate the reliable change error term based on a null hypothesis model. Once again I was confused. Only very recently did I discover that expressions 5 and 5* yield identical results when applied to the same data set (N.R. Temkin, personal communication) and when estimated variances are used (G. Maassen, personal communication). The reason for expressions 5 and 5* yielding slightly different error term values using the Temkin et al. (1999) data was due to use of nonidentical samples for parameter estimation. The use of non-identical samples came to light in the review process of the present Maassen paper—which Maassen now indicates in an author's note. Thus there were indeed only three approaches to consider (Expressions 4, 5, & 6). Nonetheless, Maassen maintains (personal communication) that Expression 5, as elaborated by Christensen and Mendoza (1986), represents random errors comprising the error distribution of a given person, whereas Expression 5* refers to the error distribution of a given sample. While it seems clear on the surface that the expressions represent separate statistical entities, it remains unclear to the present author how these expressions can then yield identical values when applied to test–retest data derived from a single normative group. Unfortunately however, my confusion does not stop there.