Description
Fit Dynamic Structural Equation Models.
Description
Applies dynamic structural equation models to time-series data with generic and simplified specification for simultaneous and lagged effects. Methods are described in Thorson et al. (2024) "Dynamic structural equation models synthesize ecosystem dynamics constrained by ecological mechanisms."
README.md
Dynamic structural equation models
Package dsem fits dynamic structural equation models, which includes as nested submodels:
- structural equation models
- vector autoregressive models
- dynamic factor analysis
- state-space autoregressive integrated moving average (ARIMA) models
The model has several advantages:
- It estimates direct, indirect, and total effects among system variables, including simultaneous and lagged effects and recursive (cyclic) dependencies
- It can estimate the cumulative outcome from press or pulse experiments or initial conditions that differ from the stationary distribution of system dynamics
- It estimates structural linkages as regression slopes while jointly imputing missing values and/or measurement errors
- It is rapidly fitted as a Gaussian Markov random field (GMRF) in a Generalized Linear Mixed Model (GLMM), with speed and asymptotics associated with each
- It allows granular control over the number of parameters (and restrictions on parameters) used to structure the covariance among variables and over time,
dsem is specifically intended as a minimal implementation, and uses standard packages to simplify input/output formatting:
- Input: time-series defined using class ts, with
NAfor missing values - Input: structural trade-offs specified using syntax defined by package sem
- Output: visualizing estimated trade-offs using igraph
- Output: access model output using standard S3-generic functions including
summary,predict,residuals,simulate, andAIC
Please see package vignettes for more details regarding syntax and features.