QuantDev Brownbag - Unsupervised Model Construction in Continuous-Time

Google map

The idiographic framework has been invaluable for understanding the dynamics of individuals as they unfold through time. However, some concerns have been levied regarding the generalizability of person-specific results to broad classes of individuals. Addressing the challenge of bridging person- and group-level inference remains a pivotal issue in quantitative methods. A broad class of models—referred to as “idio-thetic” methods—describe the spectrum between purely idiographic models to fully constrained group-level or “chained” models.

Much of the advancement in idio-thetic methods has been within the realm of discrete-time literature. Discrete-time models offer intuitive interpretations and ease of implementation; however, they are limited when compared to continuous-time models. The latter are adept at handling unequally spaced data—which frequently occur in real-world empirical applications—and possess other unique advantages. Despite this, development of novel methods in continuous-time still lags behind those in the discrete-time literature.

In my final presentation as a graduate student in QuantDev, I introduce a continuous-time extension of the group iterative multiple model estimation (GIMME) procedure. This work streamlines the fitting of continuous-time models to individual processes and leverages person-specific information to identify common, group-level structures. The discussion will cover the formal algorithm, the strengths of moving into a continuous-time framework alongside OpenMx, and present preliminary simulation results.