Kevin McKee
Virginia Commonwealth University
Kevin McKee is a predoctoral student at Virginia Commonwealth University in Richmond, VA, United States, interested in developing statistical methodology for the analysis of psychological time series data.
Background:
Cross-sectional measures of affect suffer from interpretive ambiguity and imprecision due to reliance on respondents’ retrospective abilities. Causal inferences are limited by their exclusion of temporal information. With state-based measures of affect, respondents only use short-term recall in frequently repeated assessments. Individual traits may be estimated from the statistical properties of resulting time series. If the limitations with trait-based self-assessment are ameliorated by estimating traits from repeated momentary assessments, then trait scores in the latter design should provide more valid and reliable phenotype for use in genetic studies.
A wide variety of methods exist for analyzing time series data. General tendencies may include simple means and variances within-individual. With a dynamical systems approach, a latent differential equation (LDE) is estimated with parameters that characterize the damping and frequency of oscillations around equilibrium, given random disturbances to the system. [1,2] It is demonstrated that without a method to detect and model each random disturbance or event, parameter estimates are severely biased. Simulations were run to test parameter estimation with and without a novel event-detection algorithm.
Objective:
To examine the minimum data requirements for parameter recovery with the Latent Differential Equation model, and to propose additional event-detection methods to improve parameter estimates.
Method:
Data were simulated under various conditions including different series length, signal/noise ratio, and event probability per timepoint. The model was fit to 250 randomly generated series per condition to examine the distributions of parameters estimates. The simulations were re-run with a novel event-detection algorithm.
Results:
Basic simulations of the Latent Differential Equation model were not robust to multiple disturbances or initial conditions with peaks located at any time point other than the first of the series. Analysis with event-detection corrected parameter bias under a wide variety of conditions, with degrading performance under 3:1 signal-to-noise ratio and > ~20% event probability.
[1]Boker, S., Neale, M., & Rausch, J. (2004). Latent differential equation ….. indicators. In Recent developments on structural equation models (pp. 151–174). Springer.
[2]Chow, S.-M., Ram, N., Boker, S. M., Fujita, F., & Clore, G. (2005). Emotion ….. damped oscillator model. Emotion, 5(2), 208.
Statistical Methods , Personality, Temperament, Attitudes, Politics and Religion , Positive Psychology/Wellbeing
