Focus on Psychometrics Single-Item Indicators in Nursing Research

The purpose of this article is to explore the use of single-item indicators in nursing research. Single-item indicators can be classed into two fundamentally different groups. In the first group are single-item instruments that were designed as single-item measures. These measures generally are used to obtain the subject’s perception of particular dimensions of multidimensional concepts or of an overall concept. Global single-item measures allow the subject to define the concept in a way that is personally meaningful, providing a measure that can be responsive to individual differences. Global single-item indicators require that subjects consider all aspects of a phenomenon, ignore aspects that are not relevant to their situations, and differentially weight the other aspects according to their values and ideals in order to provide a single rating. They represent a holistic way to measure subjects’ perceptions of many concepts that are of interest to nursing and are consistent with nursing’s perspective. Whenever nurse researchers are interested in individuals’ perceptions of a particular situation, perhaps in order to predict their behavior, a global single-item indicator may be a more valid measure of the concept of interest. The second group consists of single-item indicators that were designed as part of a multi-item scale. These single-item measures often are used in a particular study when the multi-item scale does not perform satisfactorily or as proxy measures for a concept that seems important in understanding the findings but that was not measured in a systematic way. The problem with this practice lies in the way multi-item measures are constructed. Items in a multi-item instrument generally are chosen so that each represents one aspect of the concept, with the goal of adequately sampling from the domain of possible items (Nunnally, 1978). Although this strategy is appropriate and important to the construction of a multi-item scale, it means that a composite of the items is necessary in order to validly measure the concept. Using one item originally designed as part of a multi-item scale may not provide a complete picture of the concept. A study of single- versus multi-item measures serves as an example of this problem. Bukowski, Ferber-Goff, and Newcomb (1990) measured antisocial behavior in school children with five items, each measuring a different behavior that is part of the larger construct. In order to compare 1-month test–retest reliability for scales made of different numbers of items, they chose one item as the single-item measure and then progressively added the other items to it. Test–retest reliability for the two- to five-item measures (r = .67 to .85) was higher than for the single-item measure (r= .55 and .69), even after correction for attenuation due to measurement error. They reason that the increased reliability of the multi-item scales is due to a factor other than the number of items included in the scale. Since scales based on more items usually have an increased range of scores which increases the correlation coefficients, other things being equal (Pedhazur, 1982), the higher reliability for the multi-item measures could be due to differences in score range and not to real differences in reliability. Bukowski et al. did not explain how they chose the single item or the order of item entry into the score. Thus, the differences in correlations also could be due to the content of the specific items and not to the number of items. Single-item measures in this second group should not be used for testing study hypotheses in planned, a priori analyses, but may be beneficial in exploring the data for suggestions for future research.