In evaluating individuals, how much of a change in scale scores should be considered a “real” change?

Answer: A rough approximation to statistical criteria can be applied to scale scores for individuals by comparing the changes with the standard error of measurement for each scale, as follows: (a) On the right-hand side of Appendix D, locate the columns headed SE of Meas; (b) if you are assessing an individual who is considered to need mental health services, look in the column headed Ref; (c) if you are assessing an individual who is not considered to need mental health services, look in the column headed Nonref; (d) look down the column until you reach the scale on which you wish to evaluate change; (e) if an individual’s score on a scale has changed more than the SE of Meas indicated in the appropriate column for the relevant scale, the change exceeds the change that is likely to occur by chance 68% of the time. Users who prefer more stringent criteria can multiply the SE of Meas by 1.65 to obtain a 90% confidence interval around the estimated true score. To obtain a more precise confidence interval, users can first estimate the individual’s true score using the following formula: (1 - rxx) X + rxxX, where rxx = the reliability of the scale, X = the mean of observed scores for the most relevant group (e.g., referred or non-referred), and X = the individual’s obtained scale score (Pedhazur & Schmelkin, 1991, p. 110). It should be remembered, however, that the standard error of measurement provides only a rough guideline for judging whether changes in an individual’s scale scores are likely to exceed chance expectations.

Comments are closed