A new scale for assessing perceptions of chance: A validation study

Background. Clinicians and researchers often wish to know how patients perc eive the likelihoods of health risks. Little work has been done to develop and validate scales and formats to measure perceptions of event probabiliti es, particularly low probabilities (i.e., <1%). Objective. To compare a new visual analog scale with three benchmarks in terms of validity and reliabi lity. Design. Survey with retest after approximately two weeks. Respondents estimated the probabilities of six events with the new scale, which featur ed a "magnifying glass" to represent probabilities between 0 and 1% on a lo garithmic scale. Participants estimated the same probabilities on three ben chmarks: two linear visual analog scales (one labeled with words, one with numbers) and a "1 in x" scale. Subjects. 100 veterans and family members an d 107 university faculty and students. Measures. For each scale, the author s assessed: 1) validity-the correlation between participants' direct rankin gs (i.e., numbering them from 1 to 6) and scale-derived rankings of the rel ative probabilities of six events; 2) test-retest reliability-the correlati on of responses from test to retest two weeks later; 3) usability (missing/ incorrect responses, participant evaluation). Results. Both the magnifier a nd the two linear scales outperformed the "1 in x" scale on all criteria. T he magnifier scale performed about as well as the two linear Visual analog scales for validity (correlation between direct and scale-derived rankings = 0.72), reliability (test-retest correlation = 0.55), and usability (2% mi ssing or incorrect responses, 65% rated it easy to use). 62% felt the magni fier scale was a "very good or good" indicator of their feelings about chan ce. The magnifier scale facilitated expression of low-probability judgments . Far example, the estimated chance of parenting sextuplets was orders of m agnitude lower on the magnifier scale (median perceived chance 10(-5)) than on its linear counterpart (10(-2)). Participants' assessments of high-prob ability events (e.g., chance of catching a cold in the next year) were not affected by the presence of the magnifier. Conclusions. The "1 in x" scale performs poorly and is very difficult for people to use. The magnifier scal e and the linear number scale are similar in validity, reliability, and usa bility. However, only the magnifier scale makes it possible to elicit perce ptions in the low-probability range (<1%).