Studying developmental processes in accelerated cohort-sequential designs with discrete- and continuous-time latent change score models.

Abstract

Studying the time-related course of psychological processes is a challenging endeavor, particularly over long developmental periods. Accelerated longitudinal designs (ALD) allow capturing such periods with a limited number of assessments in a much shorter time framework. In ALDs, participants from different cohorts are measured repeatedly but the measures provided by each participant cover only a fraction of the time range of the study. It is then assumed that the common trajectory can be studied by aggregating the information provided by the different converging cohorts. We conducted a Monte Carlo study to evaluate the practical relevance of using discrete-and continuous-time latent change score models for recovering the trajectories of a developmental process from ALD data under different sampling conditions. We focused on exponential trajectories typically found in the development of cognitive abilities from childhood to early adulthood. The results support the appropriateness of ALD designs to study such processes under various conditions of sampling. When all cohorts are drawn from the same population, both discrete-and continuous-time models are able to recover the parameters defining the underlying developmental process. However, discrete-time models yield biased estimates when time lags between observations are not constant. When cohorts are not from the same population and, thus, lack convergence, both types of models show bias in various parameters. We discuss the findings in the context of developmental methodology, encourage researchers to adopt continuous time models to analyze data from ALDs, and provide recommendations about how to implement such research designs.

Controlling for cohort effects in accelerated longitudinal designs using continuous- and discrete-time dynamic models.

“…Previous research has shown that, when the assumption of cohort equivalence is met and thus there are no cohort differences, the parameters of the generating process can be adequately recovered. For example, Estrada and Ferrer (2019) showed that, with sample sizes above 200 cases, the application of various ALD sampling schedules allowed recovering the parameters defining the population’s trajectory with a LCS model (McArdle, 2009). However, Estrada and Ferrer (2019) also found that, in the presence of cohort effects, several parameters defining key features of the population trajectory were recovered with substantial bias.…”

“…For example, Estrada and Ferrer (2019) showed that, with sample sizes above 200 cases, the application of various ALD sampling schedules allowed recovering the parameters defining the population’s trajectory with a LCS model (McArdle, 2009). However, Estrada and Ferrer (2019) also found that, in the presence of cohort effects, several parameters defining key features of the population trajectory were recovered with substantial bias. This bias led to poor confidence interval coverage rates, and was found for parameters that differed across cohorts, and importantly, also for parameters that were cohort-invariant.…”

“…Therefore, a CT model is proposed as a more theoretically accurate representation of those processes, as it explicitly captures this feature (Oud & Delsing, 2010). Fourth, the LCS-DT model described in the previous section is a specific case of an underlying CT model (Estrada & Ferrer, 2019). Such CT model contains the same information as the DT model and more.…”

Handling Time in Occupational Stress and Well-Being Research: Considerations, Examples, and Recommendations

“…Many of the analytical methods discussed earlier with respect to limitations may well be appropriate for certain research projects. Researchers need not limit themselves to traditional longitudinal panel data, either: there are alternative collection modes, many of which may better "tap into" our phenomena of interest, that bear consideration in future research (e.g., accelerated cohort designs, intensive longitudinal design; see CTSEM discussion of such data in Estrada & Ferrer, 2019;Ryan et al, 2018). Moreover, the advent of continuous time meta-analysis (Dormann et al, 2019;Guthier et al, 2020) is an exciting recent development in this literature, and it stands to contribute to time-cognizant occupational health research designs, analyses, and practical applications moving forward.…”

Modeling Retest Effects in Developmental Processes Using Latent Change Score Models

“…This paper focuses on modeling retest in latent change score (LCS) models (McArdle, 2009). LCS models were developed explicitly to model data that follow exponential forms, and have been extensively used in the analyses of developmental data, especially cognitive abilities (Estrada et al, 2020;Estrada & Ferrer, 2019;Ferrer et al, 2007;Ferrer et al, 2010;Ghisletta & Lindenberger, 2003, 2005Grimm, 2008;Kievit et al, 2018;among others). LCS models offer a general framework, of which growth models such as a latent growth curve can be a special case.…”