This package has been designed in order to provide various psychometric network modelling techniques for multivariate time series data from the behavioral domain, in a simple wrapper function. The aim is to estimate the temporal and contemporaneous interaction structure in a (possibly) sparse fashion and thereby being able to visualize these interactions by means of conditional independence graphs. Depending on the model, the interactions are estimated at the individual level -- intra-individual dynamics -- or population level -- inter-individual dynamics.
In the last decade, time series data became popular in the behavioral sciences. These data, allow us to study the dynamics of complex behavioral systems both at the individual and the population level. Vector autoregression (VAR) is the cornerstone in the statistical modelling of multivariate time series data and various VAR extensions are available that can handle cases where additional complexities are imposed in the analysis.
This package introduces the main psychNET function that is used to fit various dynamic models. Models that can be fitted to time-series data from one person using the psychNET function are:
Vector Autoregressive (VAR) models: Traditional VAR model or VAR model with exogenous variables (VARX). See the help file from the VAR() function from the R package vars for more details.
Sparse Vector Autoregressive (SVAR) models: SVAR model estimated using penalized multivariate least squares (see fitVAR function in the R package sparsevar). SVARs estimated using penalized least squares with simultaneous lag estimation (SVARHL) and exogenous variables (SVARHLX) implemented in the functions sparseVAR and sparseVARX respectively from the R package bigtime. SVAR model with simultaneous covariance estimation using penalized likelihood approach known as graphical VAR (GVAR) implemented in the function graphicalVAR from the package graphicalVAR. A SVAR model for mixed type of time series with marginal distributions in the exponential family (SMVAR), implemented in the function mvar from the R package mgm.
Sparse Vector Error Correction Model (SVECM): SVECM estimated using penalized multivariate least squares that is implemented in the function fitVECM from the R package sparsevar.
Sparse Vector Autoregressive Moving Average (SVARMA) model: SVARMA model where the parameters are estimated using penalized least squares with simultaneous lag estimation (SVARMAHL) implemented in the function sparseVARMA from the R package bigtime.
Sparse Time-Varying Vector Autoregressive (TV-SVAR) model: TV-SVAR model where the parameters are allowed to vary smoothly with time. This model can handle time series with marginal distributions in the exponential family. The model is implemented in the function tvmvar, which is available with the R package mgm. (This model is NOT yet available in the psychNET package. To be expected with the next update.)
Dynamic factor model (DFM): DFM with three different types of estimates for the latent series. For details you can check the function dfm from the R package dynfactoR available at github.
For network inference at the population level from nested time-series data, two models in the broad class of VAR models can be fitted via the psychNET function. These are:
Multilevel Vector Autoregressive (MLVAR) model: This model estimates group as well as individual level effects and it is implemented in the R function mlVAR from the R package mlVAR.
Sparse Vector Autoregressive (SVAR) model: This model estimates sparse group level networks from nested time series with simultaneous covariance estimation. We call this model group graphical VAR (GGVAR) and it has been primarily implemented in the function sparse.tscgm available with R package SparseTSCGM.
Lutkepohl, H. (2006). New Introduction to Multiple Time Series Analysis. Springer, New York.
Basu, S., Michailidis, G. (2015). Regularized estimation in sparse high-dimensional time series models. Ann. Statist. 43, no. 4, 1535-1567.
Abegaz, F., Wit, E. (2013). Sparse time series chain graphical models for reconstructing genetic networks. Biostatistics. 14, 3: 586-599.
Haslbeck, J., Waldorp, L. J. (2016). mgm: Structure Estimation for time-varying Mixed Graphical Models in high-dimensional Data.
Nicholson, W. B., Bien, J., Matteson, D. S. (2017). High Dimensional Forecasting via Interpretable Vector Autoregression..
Wilms, I., Basu, S., Bien, J., Matteson D. S. (2017). Sparse Identification and Estimation of High-Dimensional Vector AutoRegressive Moving Averages.
Epskamp, S., Waldorp, L. J., Mottus, R., Borsboom, D. (2016). The Gaussian Graphical Model in Cross-sectional and Time-series Data.
# NOT RUN {
# Load the R package psychNET
library(psychNET)
# }
Run the code above in your browser using DataLab