Modeling intraindividual variability as a predictor with intensive longitudinal data

In many areas of psychology, researchers are interested in studying whether intraindividual variability (IIV) is predictive of behavioral and health outcomes after controlling for the intraindividual mean. To the end, IIV indicators such as the observed intraindividual variance (OIVAR) or the observed intraindividual standard deviation (OISD) are often modeled as a predictor in regular regression analysis. However, OIVAR and OISD have been found to have a low-reliability problem, especially when the number of occasions is small. In this study, we analytically examined statistical features (mean and variance) of OIVAR and OISD. The results revealed that when measurement errors exist, regular regression can yield (a) more accurate results for the coefficient of intraindividual variance (IVAR) but (b) worse results for the coefficient of intraindividual standard deviation (ISD) when the number of occasions increases. Furthermore, we compared the performance of alternative modeling approaches, including the time-parceling, single indicator latent variable, and Bayesian variability modeling approaches, to that of regular regression for modeling IVAR or ISD as a predictor. Simulation results were consistent with the analytical results and further suggested that our proposed (a) time-parceling with bootstrapping and (b) Bayesian variability modeling approaches performed well and better than regression for modeling IVAR as a predictor. When measurement errors exist, only the proposed Bayesian variability modeling approach performed well for modeling ISD as a predictor. Implications of the results and recommendations were discussed. (PsycInfo Database Record (c) 2026 APA, all rights reserved).