Regularization-based graphical modeling has key drawbacks, including difficulty obtaining valid parameter uncertainty and inflated false positive rates¶

Claim · All claims
Evidence · 1 study · design r1 · q1
Currency · newest study 2021 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · overdue for a recheck since 2024: tool and model accuracy changes quickly

Subclaims¶

q1 i? Obtaining a valid measure of parameter uncertainty is very difficult with regularization-based graphical modeling. → Donald Williams 2021 q1 i? Regularization can produce an inflated false positive rate in graphical modeling. → Donald Williams 2021

Evidence¶

Donald Williams 2021¶

Donald Williams. (2021). GGMnonreg: Non-regularized Gaussian graphical models in R. Journal of Open Source Software, 6(67), 3308. https://doi.org/10.21105/joss.03308

q1 · i? · design · r1

The article asserts, citing prior literature (Bühlmann et al., 2014; Donald R. Williams et al., 2019), that "there can be an inflated false positive rate" and that valid parameter uncertainty is very difficult to obtain under regularization. No new empirical test of these drawbacks is reported in this article.

"There are key drawbacks of regularization, including, but not limited to, the fact that obtaining a valid measure of parameter uncertainty is very (very) difficult (Bühlmann et al., 2014) and there can be an inflated false positive rate"

Discussion¶

Related Claims¶

• The frequentist graphical LASSO has difficulty estimating centrality indices and uncertainty in these measures in symptom networks — a narrower finding that bears on this claim
• Non-regularized graphical model methods are specifically designed for low-dimensional data common in the social-behavioral sciences — related
• Bayesian GLASSO shows good coverage of uncertainty for strength and closeness centrality, but uncertainty in betweenness centrality is estimated less well — a narrower finding that bears on this claim
• In simulations, Bayesian GLASSO estimation outperforms frequentist GLASSO on bias in edge weights, centrality measures, correlation between estimated and true partial correlations, and specificity — a narrower finding that bears on this claim
• Sensitivity is better for the frequentist GLASSO than for the Bayesian GLASSO, though Bayesian GLASSO performance is usually close — related
• In simulations, the Bayesian GLASSO performed better than the Horseshoe prior for estimating symptom networks — related