Within-Subject Precision Measurement: Theory and Utility of Assessing Change Scores and Their Variance
En palabras de los autores
Within-subject studies gain statistical power by removing stable between-subject differences from the comparison. However, the metrics commonly used to select measurement instruments and anticipate their within-subject performance are often based on between-subject correlations of raw scores from test-retest designs. Measuring each subject's change directly -- after an intervention, and ideally more than once -- provides a firmer basis for evaluating measurement precision and planning studies of within-subject change. Here we describe the statistical theory behind estimating measurement properties from change scores. In one instantiation, the pre- and post-intervention measurements are each made twice, yielding the duplicate change-score design. Such a design provides an estimate of the mean signal change and enables the decomposition of the total observed variance of change scores into true between-subject response heterogeneity and measurement noise. In extended designs, where multiple measures are collected within and across days, measurement noise can be further decomposed to distinguish within-session from between-day measurement variability. These estimates supply the inputs needed for within-subject study design, including providing a basis for selecting among alternative measurements and comparing processing pipelines, performing power calculations, and estimating the precision limits of detecting change within individual subjects.
Apareció: jueves, 19 de noviembre. bioRxiv. Preprint, todavía sin revisión por pares.