Analytically, the DR estimator's bias is bounded by the product of propensity and imputation errors, and its generalization bound contains a variance term scaling with squared imputation error¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

Reading the evidence codes

q is the design tier, i the impact, n the sample, and the kind with its r the rigour judged within that kind: what each code means.

This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.

q2 i? Lemma 1 proves the DR estimator is unbiased if either the propensity model or the imputation model is accurate, with bias bounded by the product of the two models' errors. → Peilin Zhan 2026 q2 i? Theorem 1 derives a generalization bound whose variance term scales with squared imputation error, motivating temporal smoothness to tighten the bound. → Peilin Zhan 2026 (2)

Evidence¶

Peilin Zhan 2026¶

Peilin Zhan, Wei Chen, Weilin Chen, Shuyi Pan, Ruichu Cai. (2026). Temporal Smoothness Doubly Robust Learning for Debiased Knowledge Tracing. https://arxiv.org/abs/2605.05958 doi:10.48550/arxiv.2605.05958

q2 · i? · design · r2

Analytical derivation (Lemma 1, Section 4.3) with proof sketch in the article and full proofs in Appendix F. The article states the bias bound confirms the double robustness property and yields second-order error decay when both models are learned with expressive estimators.

"The bias of the DR estimator is bounded by the product of the errors of the two auxiliary models"

Peilin Zhan 2026 (2)¶

Peilin Zhan, Wei Chen, Weilin Chen, Shuyi Pan, Ruichu Cai. (2026). Temporal Smoothness Doubly Robust Learning for Debiased Knowledge Tracing. https://arxiv.org/abs/2605.05958

q2 · i? · design · r2

Analytical derivation (Theorem 1, Section 4.4) using the Azuma-Hoeffding inequality over the sequential interaction structure. The article states inaccuracy in the imputation model loosens the bound and leads to unstable training, motivating the temporal smoothness constraint.

"the total generalization error is dominated not just by the bias product but also by the Variance Term, which scales with the squared imputation errorδ 2 t,c"

Discussion¶

Retrieval-augmented generation reduces but does not eliminate hallucinations in AI-assisted verification — related

Page metadata

Field	Value
`type`	`claim`
`id`	`dr-bias-product-and-variance-bound`
`title`	`Analytically, the DR estimator's bias is bounded by the product of propensity and imputation errors, and its generalization bound contains a variance term scaling with squared imputation error`
`description`	`Analytically, the DR estimator's bias is bounded by the product of propensity and imputation errors, and its generalization bound contains a variance term scaling with squared imputation error`
`status`	`review`
`evidence_strength`	`weak`
`newest_evidence`	`2026`
`generated`	by `process:wiki-ingest` on `2026-10-11`
`trust`	unverified — no one has confirmed this page's content
`sources`	2 — `peilin-zhan-2026`, `peilin-zhan-2026-2`

Frontmatter is the page's machine-readable record — see the Schema & Guide.