Latent State Trait Modelling of Executive Functions
Personality Psychology, Clinical and Medical Psychology, Methodology
Final Report Abstract
Executive functions (EFs) are general-purpose, higher-order control processes that serve to coordinate more basic cognitive, perceptual and motor processes. EFs are thought to comprise the dimensions of inhibition, updating and shifting. In addition to a long tradition of experimental studies, EFs have increasingly been investigated with regards to individual differences, using statistical approaches including correlations, multiple regression and structural equation modelling (SEM). Accordingly, the question of temporal stability and reliability has gained in importance, and a number of studies have reported that measures of EF have poor test-retest reliabilities. Previous reliability studies are however limited in that they have not formally modelled the state and trait components that underlie variance in EF task performance. Such modelling can be performed within the framework of latent state-trait (LST) theory, which allows the estimation of the amounts of variance in a measurement due to stable trait, situational fluctuations (state) and person x situation interactions, and measurement error. LST has previously been applied in psychometric and biopsychological research, but no study has utilised this approach with regards to EFs. Therefore, this project aimed to provide a detailed investigation of the test-retest reliability and internal consistency as well as the state and trait components of a comprehensive battery of EF task variables in a large sample. The sample consisted of N=289 young, healthy adults, of whom N=250 completed all sessions. Four sessions per participant were carried out at weekly intervals, with time of day kept consistent within each participant across sessions. The EF battery included the go/nogo and stop-signal tasks for assessment of inhibition, the n-back and tone-monitoring tasks for updating, and the number-letter and local-global tasks for shifting. The results showed that LST model fits were at least acceptable for most variables. It was also observed that LSTT models, which do not allow for systematic change in trait ability over time, did not provide a better fit than models that do allow such change (LST) or explicitly model it (LGC). Intraclass coefficients (ICC) were at least good for EF variables, except some measures from switching and updating tasks. Cronbach’s alphas were generally very good. Estimates from LST models indicated very good reliabilities for most variables, and the majority of reliable variance could be attributed to trait influences. Overall, the results from this study show that most measures of EFs have good reliabilities and that trait influences play a major role in interindividual variance in performance.
