Modeling intraindividual means and variances from ecological momentary assessment data: comparing standard computational formulas to mixed-effects location-scale model estimates.

Traditionally, intraindividual means and variances derived from ecological momentary assessment (EMA) data have been calculated using standard computational formulas (SCF), such as subject-level means and standard deviations. However, these SCF methods assume uniform precision across subjects, disregarding variation in the number of observations, missing data issues, and the non-continuous nature of data scales. This study evaluated the predictive accuracy of the coefficients of intraindividual means and variances computed via SCF against those estimated using random effects from a Mixed-Effects Location Scale (MELS) model. A five-scenario simulation study was conducted: (1) varying numbers of observations per subject, (2) varying mean-to-variance ratios, (3) varying proportions of missing data under a missing completely at random (MCAR) assumption, (4) varying proportions of missing data under a missing at random (MAR) assumption, and (5) varying categories of ordinal scale responses. Bias and coverage of the mean levels and variability coefficients were compared across methods. In addition, a real-life dataset was used to compare the difference of means and variances between SCF and MELS approaches. Results consistently showed that the MELS model approach outperformed SCF method, yielding lower bias and higher coverage of the coefficients across all scenarios. These findings support the use of MELS for more accurate and reliable estimation of intraindividual means and variances in EMA data, highlighting its advantages for subsequent predictive modeling.