Common problems in using, modifying, and reporting on classic measurement instruments.
Journal - Arthritis care and research : the official journal of the Arthritis Health Professions Association (UNITED STATES )
Users of classic summated scales can clear up some common quandaries by considering the underlying assumptions that a scale's items represent a random sample drawn from an infinite pool of items representing a unidimensional domain. Scale reliability is higher with more items and a higher average correlation among items. Standards of reliability need to be higher when decisions about treatment are to be based on scores, because confidence intervals around individual scores expand quickly below alpha = 0.95. Scales that are longer than needed may sometimes be shortened, using a formula to determine the approximate number of questions that will yield the desired reliability. Reliability may be improved by addition of new items or expansion of response sets for existing items. Such alterations of scales should always be corroborated with new data. The use of coefficient alpha for scales with severely limited domains, such as self-care knowledge, is rarely useful or appropriate psychometrically, and can be misleading. Standard item selection procedures in classic scales maximize reliability at the scale center and overestimate change at the center and underestimate it at the ends; there are both classic and modern techniques to improve scale construction in this regard. The omission of specific items (such as symptoms) on many scales is often immaterial to the reliability of the scale and to the usefulness of the summary score; this is because each item is well correlated with other scale items, so that information about the underlying domain from any single item is redundant. Finally, despite the attractiveness of disease-specific scales, generic scales often do as well, even in arthritis populations, and have the added benefit of allowing comparisons with a wider range of disease and cultural groups. The development, modification, and use of classic summated scales can be complex in practice, calling on both statistical skills and content area judgment, but at heart the principles are fairly simple. Keeping the principles in mind is half the battle when solving many common problems. The interested reader is referred to the texts referenced in this article and the one by DeVellis (1) for further discourse on this topic.
|ISSN : ||0893-7524|
|Mesh Heading : ||Activities of Daily Living Arthritis Confidence Intervals Humans Outcome Assessment (Health Care) Psychiatric Status Rating Scales Psychometrics Reproducibility of Results physiopathology psychology|
|Mesh Heading Relevant : ||Health Status Indicators|