Written by Mohammed Saqr and Sonsoles López-Pernas
Provides clean-room implementations for estimating psychometric network models, including correlation and partial-correlation networks, Gaussian graphical models with extended Bayesian information criterion (EBIC) regularization, nonparanormal and stepwise selection variants, information-filtering networks (the triangulated maximally filtered graph and the local-global inverse covariance), relative-importance networks, and Ising and mixed graphical models (doi:10.3758/s13428-017-0862-1; doi:10.1007/978-3-031-54464-4_19). All methods are implemented from first principles in base R without compiled dependencies and return consistent, tidy outputs. Functions are designed to be transparent and report optimization diagnostics where applicable. For Gaussian graphical models, the graphical lasso stationarity (Karush-Kuhn-Tucker) residual quantifies the deviation of the estimated solution from the optimum of the corresponding convex optimization problem.
psychnets estimates psychometric networks and assesses
how far they can be trusted. A psychometric network represents variables
such as questionnaire items, symptoms or behaviours as nodes and the
associations between them as edges; in a partial-correlation network an
edge is the association between two variables that remains once all the
others are taken into account (Epskamp & Fried, 2018). The network
shows which variables are directly related, which hold a central
position, and how the structure differs between groups. The package
estimates Gaussian graphical models (regularized, unregularized and
nonparanormal), information-filtering and relative-importance networks,
Ising models for binary items, mixed graphical models, and the panel
graphical VAR for a few waves of panel data. Each network can be
bootstrapped, tested for the stability of its centralities and the
reliability of its edges, and compared between groups.
Every estimator is written in base R. A regularized fit reports how
far it is from the optimum of its objective
(certificate()), so the network can be checked without a
second solver.
Vignettes
Guides
# from CRAN
install.packages("psychnets")
# development version
# install.packages("pak")
pak::pak("mohsaqr/psychnets")psychnets estimates an EBIC-regularized Gaussian
graphical model from data or a correlation matrix and reports its
certificate on printing. The example uses SRL_GPT, one of
the self-regulated-learning datasets shipped with the package (five MSLQ
constructs).
library(psychnets)
net <- ebic_glasso(SRL_GPT) # EBIC graphical lasso of the five MSLQ constructs
net # edges, lambda, gamma, and the KKT residual
#> <psychnet> glasso network
#> nodes: 5 edges: 10 (undirected)
#> lambda: 0.00861 gamma: 0.5
#> optimality (KKT residual): 2.21e-10
net_centralities(net) # strength and expected influence, tidy
#> node strength expected_influence
#> 1 CSU 1.1984512 1.19845117
#> 2 IV 1.0012765 1.00127649
#> 3 SE 0.8492185 0.84921854
#> 4 SR 1.2314317 0.53172852
#> 5 TA 0.6082238 -0.09147935The network is drawn through cograph::splot():
cograph::splot(net, layout = "spring", edge_labels = FALSE, legend = TRUE)
The certificate is available directly, so the distance of the fit from the unique optimum of its objective can be read without an external reference:
certificate(net)
#> method certificate kind certified
#> 1 glasso 2.213347e-10 kkt TRUEThe estimators, each fitted in base R; the regularized ones report a
certificate (certificate()):
| Verb | Model | Data |
|---|---|---|
cor_network() |
correlation network (with significance thresholding) | continuous |
pcor_network() |
partial-correlation network | continuous |
ebic_glasso() |
EBIC-regularized Gaussian graphical model (graphical lasso) | continuous |
huge_network() |
nonparanormal Gaussian graphical model | continuous |
ggm_modselect() |
unregularized stepwise Gaussian graphical model | continuous |
tmfg_network() |
Triangulated Maximally Filtered Graph | continuous |
logo_network() |
Local–Global sparse inverse covariance | continuous |
relimp_network() |
relative-importance network (LMG / Shapley) | continuous |
ising_fit() |
Ising model, L1-penalized | binary |
ising_sampler() |
Ising model, unregularized | binary |
mgm_fit() |
mixed graphical model | Gaussian + binary; multi-level categorical with
native = FALSE |
panel_gvar() |
panel graphical VAR: temporal, contemporaneous, and between-person networks (stationary random-intercept VAR(1), maximum likelihood; every edge with SE and p-value by default; optional Wald pruning + step-up, or EBIC lasso) | continuous panel data, 3+ waves |
psychnet() |
unified front door routing the cross-sectional estimators above | — |
The framework verbs operate on any fitted network, and on event logs
and grouped data through psychnet():
| Verb | Purpose |
|---|---|
net_centralities(), net_bridge(),
net_edge_betweenness() |
node, bridge, and edge centrality |
net_clustering(), net_smallworld() |
weighted clustering coefficients and the small-world index |
net_predict() |
node predictability (variance explained; classification accuracy) |
net_boot(), difference_test() |
bootstrapped accuracy and within-network difference tests |
net_stability(),
net_casedrop_reliability(),
net_split_reliability() |
case-dropping stability and split-half reliability |
net_compare() |
the permutation Network Comparison Test |
redundancy(), net_aggregate() |
redundant-node detection and community aggregation |
psychnet() additionally constructs networks directly
from event logs — one row per action — computing action
frequencies and separating within- from between-actor variation, and
estimates a separate network per group on request.
Every fitted network inherits the cograph plotting classes,
and each result object carries a base-graphics
plot() method.
Node centrality plots directly from its result object:
plot(net_centralities(net))
net_boot() resamples the data, re-estimates, and returns
bootstrapped edge-weight confidence intervals; edges whose interval
excludes zero are drawn in the accent colour:
set.seed(1)
plot(net_boot(SRL_GPT, method = "glasso", n_boot = 500), type = "edges")
Every estimator is written in base R. The only package dependencies are stats, parallel, graphics and grDevices; the compiled solvers of glasso and glmnet, which qgraph, bootnet, IsingFit and mgm call, are not used.
A fitted network carries its own check. For a Gaussian graphical
model, certificate() reports (and glasso_kkt()
computes) the stationarity residual of the graphical-lasso objective;
because that objective is strictly convex its minimizer is unique, so a
near-zero residual shows that the network is the optimum, with no
reference solver involved. The nodewise Ising and mixed graphical models
are checked the same way through the penalized-likelihood score
residual. Because each fit is solved to stationarity before the EBIC is
computed, no edge is selected only because the optimization stopped
early. A network changed after fitting, by a threshold or by pruning,
reports NA, and printing it shows the residual of the fit
before the change. External packages (qgraph,
IsingFit, mgm, bootnet) serve as
cross-checks, not as the definition of a correct fit.
psychnets is checked against the established reference
packages in two ways, both reproducible from a fixed set of package
versions. For the EBIC graphical lasso, the comparison limit is set by
the convergence criterion of the glasso implementation used by
qgraph, about 1e-4; psychnets solves its own
optimization to tighter tolerances.
Each model is fitted by psychnets and its reference
package on the same data. The table reports the observed comparison
criterion; it is not a claim that every independent implementation
returns numerically identical weights.
psychnets estimator |
Reference package | Data | Observed difference |
|---|---|---|---|
ebic_glasso() |
qgraph::EBICglasso 1.9.8 |
AR(1) chain, n = 1000, p = 8 | edge set identical; max |Δ| = 2.2e-06; KKT = 1.5e-10 |
ising_fit() |
IsingFit::IsingFit 0.4 |
AR(1) latent, n = 2000, p = 6, binarized | edge set identical; max |Δ| = 4.6e-03; threshold r = 1.000 |
mgm_fit() |
mgm::mgm 1.2.15 |
mixed Gaussian + binary, n = 3000, p = 4 | edge set identical; max |Δ| = 1.6e-03; KKT = 1.1e-09 |
panel_gvar() |
psychonetrics::panelgvar 0.17.8 (+ prune;
stepup supports also identical) |
stationary panel, n = 400, p = 3, 4 waves | pruned edge set identical; max |Δ| = 1.4e-04 (psychonetrics’ stopping tolerance: its gradient 6.8e-04, ours 3.0e-13); psychnets reaches a lower ML objective (by 5.7e-08) |
cor_auto() |
psych::polychoric / qgraph::cor_auto |
four-level Likert, n = 1000, p = 4 | max |Δ| = 2.0e-05 (psych), 1.7e-07 (qgraph) |
.pbivnorm() (bivariate normal CDF) |
mvtnorm::pmvnorm 1.2 |
five (h, k, ρ) points | max |Δ| = 5.6e-17 |
On real questionnaire and ability data the psychnets
EBIC graphical lasso reproduces the qgraph edge set exactly
— structure agreement equals 1 on every instrument, with a largest
edge-weight difference of 0.008 — and the Ising model matches
IsingFit on binary batteries. The instruments, all loaded
from installed CRAN packages so the comparison downloads nothing:
| Instrument | Source | Items | n | Reference | max |Δ| |
|---|---|---|---|---|---|
qgraph::big5 |
Dolan, Oort, Stoel & Wicherts (2009) | 240 | 500 | qgraph | 3.6e-05 |
psych::bfi |
Revelle, Wilt & Rosenthal (2010) | 25 | 2760 | qgraph | 3e-06 |
psych::sat.act |
Revelle (psych) | 6 | 700 | qgraph | 2e-06 |
psychTools::msq |
Revelle & Anderson (1998) | 50 | 3876 | qgraph | 1.2e-05 |
psychTools::epi.bfi |
Eysenck & Eysenck (1964) | 13 | 231 | qgraph | 8.0e-03 |
psychTools::epi |
Eysenck Personality Inventory | 50 | 3476 | qgraph | 6e-06 |
psychTools::sai |
Spielberger State Anxiety | 22 | 5282 | qgraph | 3.5e-05 |
psychTools::tai |
Spielberger Trait Anxiety | 21 | 3016 | qgraph | 7e-06 |
psychTools::spi |
Condon (2018), SAPA Inventory | 50 | 4000 | qgraph | 1.6e-05 |
psychTools::blot |
Bond’s Logical Operations Test | 35 | 150 | qgraph | 0 |
mgm::Fried2015 |
Fried et al. (2015) | 11 | 515 | qgraph | 8e-06 |
mgm::PTSD_data |
McNally et al. (2015) | 6 | 344 | qgraph | 1.9e-05 |
mgm::B5MS |
Haslbeck & Waldorp (mgm) | 5 | 500 | qgraph | 0 |
mgm::symptom_data |
Haslbeck & Waldorp (mgm) | 48 | 1476 | qgraph | 1.5e-05 |
NetworkToolbox::neoOpen |
Christensen, Cotter & Silvia (2019) | 48 | 802 | qgraph | 4e-06 |
networktools::depression |
Jones (networktools) | 9 | 1000 | qgraph | 6.5e-05 |
networktools::social |
Jones (networktools) | 16 | 350 | qgraph | 7e-06 |
EGAnet::optimism |
Golino & Christensen (EGAnet) | 10 | 272 | qgraph | 6e-06 |
EGAnet::intelligenceBattery |
Golino & Christensen (EGAnet) | 50 | 1152 | qgraph | 2.2e-05 |
psychTools::ability |
Revelle; ICAR ability items | 16 | 1248 | IsingFit | 5.9e-02 |
psychTools::iqitems |
Condon & Revelle; ICAR | 11 | 1523 | IsingFit | 1.2e-02 |
Where the truth is known — data drawn from a known sparse precision
matrix — the psychnets and qgraph estimators
recover the generating structure at the same F-measure. The full
derivation, the worked graphical-lasso case, and the complete tables are
given in the package’s technical report.
cograph.