Putting affective dynamics into context: Towards a realistic and reliable modeling approach
General, Cognitive and Mathematical Psychology
Final Report Abstract
With the project “Putting affective dynamics into context: Towards a realistic and reliable modeling approach”, we set out to address and balance two important challenges of the prevalent dynamic paradigm of affect research: A need for more realistic modeling approaches to gain deeper insights into affective functioning and a need for more reliable dynamic modeling solutions to ensure the trustworthiness of these insights. A trade-off between both challenges arises as typical psychological data are limited, constraining the amount of information available to obtain more complex modeling solutions reliably. We focus on the popular class of (vector) autoregressive models, whose parameters capture temporal dependencies in one or among multiple affective processes and are therefore typically used to shed light on the regularities of affective functioning. One way to render (vector) autoregressive models more realistic is to enrich them with contextual variables that moderate single to all model parameters thus allowing to investigate context-tied changes in affect dynamics. Such models might help to address notions of situation-specific emotion regulation and to animate abstract ideas of (mal-)adaptive development. We also consider other model extensions, such as the increasingly popular continuous-time formulations. These provide more realistic accounts of data from typical intensive longitudinal studies, which often employ flexible sampling schemes. Reliability of modeling solutions is mainly cast in terms of estimation reliability and hence the precision (i.e., standard errors) of individual and the distinguishability (i.e., correlation) of different parameter estimators. Estimation reliability is an important foundation for inference and the interpretability of model parameters. We complement this at times by looking at predictive accuracy, assessing reliability of modeling solutions as a whole with an emphasis on their generalizability to unseen data. In a first subproject, we investigate how well continuous-time vector autoregressive models can be estimated from data sampled at different rates. We examine the relationship between estimation reliability and sampling rates on the basis of maximum likelihood estimation theory. We show how sampling rates matter and that reliability-optimal sampling rates depend on the dynamics of the process under study. We thus provide results that can be used to plan the sampling rates of future studies. This includes analytical results as well as open-source functions in the R environment. In a second subproject, we consider autoregressive models with observed contextual covariates (i.e., intercept-moderated autoregressive models). We show analytically and in simulations how temporal dependence in the covariate alone can lead to heightened predictor correlations, thus constituting a special instance of the well-known collinearity problem in classical regression modeling. This results in impaired estimation reliability, specifically trade-offs between the predictors’ regression coefficient estimators. We advise to account for the expected temporal dependence in contextual covariates during study planning (by e.g. reducing it if controllable or gathering sufficient time points overall). In a third, ongoing, subproject, we still focus on autoregressive models with contextual covariates, but assume that contextual changes happen rarely, at unknown times, and impact the formalized affective process substantially but in unknown ways. We show under which conditions such contextual contamination can leverage affective dynamics and consequently their estimation reliability and predictive accuracy. A main ingredient thereby is the use of robust autoregressive modeling. The subproject comprises a comprehensive simulation study.
Publications
- (2020, October). Time series analysis of intensive longitudinal data in psychosomatic research: A methodological overview. Journal of Psychosomatic Research, 137, 110191, 1-15
Ariens, S., Ceulemans, E., & Adolf, J. K.
(See online at https://doi.org/10.1016/j.jpsychores.2020.110191) - (2021). Selection of the number of participants in intensive longitudinal studies: A user-friendly Shiny app and tutorial for performing power analysis in multilevel regression models that account for temporal dependencies. Advances in Methods and Practices in Psychological Science, 4(1), 1-24
Lafit, G., Adolf, J. K., Dejonckheere, E., Myin-Germeys, I., Viechtbauer, W., & Ceulemans, E.
(See online at https://doi.org/10.1177/2515245920978738) - (2021, July 30). Collinearity issues in autoregressive models with time-varying serially dependent covariates
Ariens, S., Adolf, J., & Ceulemans, E.
(See online at https://doi.org/10.31234/osf.io/96snh) - (2021, June 24). Optimal Sampling Rates for Reliable Continuous-Time First-Order Autoregressive and Vector Autoregressive Modeling. Psychological Methods
Adolf, J. K., Loossens, T., Tuerlinckx, F., & Ceulemans, E.
(See online at https://doi.org/10.1037/met0000398)
