Psychometric Properties of Instruments
Because classification in psychiatric epidemiology is based on self-report or interviewer-based rating scales and questionnaires, several features of an instrument must be tested and quantified. An instrument’s sensitivity (proportion of those with true-posi-tive results identified as such by the study instrument) and spe-cificity (proportion of those with true-negative results identified as such by the study instrument) are one set of measures. For an instrument to be useful in epidemiology, it should have high sen-sitivity and at least moderately high specificity.
To identify cases accurately, an instrument used for case identification must be reliable and valid. Reliability refers to the reproducibility of a measure (i.e., the consistency of measurement regardless of the rater, the situation, or the time of administration). Interrater agreement is usually calculated with statistical meth-ods, such as the kappa statistic, that control for chance agreement. Test–retest reliability, or temporal stability, is calculated with product-moment or intraclass correlation coefficients. Validity refers to whether a construct is measured accurately. This concept is more difficult to establish in psychiatry because there is no “gold standard” or biological marker for the disorders under study.