I’m reaching out for ideas to improve performance issues related to quasi-complete separation encountered in about 1% of my datasets. The majority of the datasets work fine, but this small subset poses a challenge. Model Overview: Standard logistic regression, utilizing QR decomposition for ortho…

In this example, the model “knows” with very high precision what value the linear predictor should take when the covariate is -0.5, and also “knows” that the slope should be large and positive, but is very uncertain about just how large the slope should be. You can see intuitively why this would in…

Thank you! The intuition you provided here is very helpful. I should be able to come up with a reasonable set of heuristics based on this.

Seeking Insights on Addressing Quasi-Complete Separation in Logistic Regression Models in Stan

Modeling

Topic		Replies	Views
Separable logistic regression performance optimizations Modeling fitting-issues , performance , logistic-regression	2	180	June 25, 2024
Obvious way to speed up huge logisitic regression? Modeling	8	723	May 24, 2018
Bayesian logistic regression workflow when facing complete separation General prior-choice , logistic-regression , model-selection	3	375	August 15, 2025
Sampler getting stuck? Modeling	18	3938	May 26, 2019
QR decomposition and non-linear models General specification	7	1093	February 12, 2021

Seeking Insights on Addressing Quasi-Complete Separation in Logistic Regression Models in Stan

Related topics