pith. sign in

arxiv: 2607.02327 · v1 · pith:XFY2D6J5new · submitted 2026-07-02 · 📊 stat.AP

Instrumented difference-in-differences under case-control sampling

Tran Trong Khoi Le , Emilie Sbidian , Tat-Thang Vo This is my paper

Pith reviewed 2026-07-03 02:39 UTC · model grok-4.3

classification 📊 stat.AP
keywords instrumental variablesdifference-in-differencescase-control samplingselection biasstructural mean modelscausal inferenceepidemiologyretrospective data
0
0 comments X

The pith

A new instrumented difference-in-differences estimator lets instruments with time-invariant direct effects serve as valid tools on the trend scale in case-control data after correcting sampling bias.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces an instrumented difference-in-differences method designed specifically for case-control studies of rare outcomes. It demonstrates that an instrument candidate whose direct effect on the outcome stays constant over time can still identify causal effects on the trend, provided the bias from retrospective sampling is modeled inside a structural mean model. Simulations confirm the estimator performs well in finite samples, and the method is applied to assess infection risk from biologic treatments for psoriasis using national claims data.

Core claim

The proposed iDiD approach accommodates IV candidates that have time-invariant direct effect on the outcome. When retrospective case-control datasets are collected, the candidate can still be used as a valid instrument on the trend scale when selection bias induced by retrospective sampling is efficiently taken into account.

What carries the argument

The iDiD estimator inside the structural mean model framework that isolates trend effects and corrects for retrospective selection bias.

If this is right

  • Researchers gain a wider pool of usable instruments in retrospective case-control studies of rare diseases.
  • Causal effects on outcome trends can be identified even when the classic exclusion restriction fails in a time-invariant manner.
  • The same correction for sampling bias applies directly to national claims databases without requiring prospective collection.
  • Finite-sample bias and coverage remain controlled under the stated modeling assumptions in the reported simulations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same modeling step could be combined with other time-series or longitudinal corrections to handle multiple sources of bias simultaneously.
  • If many real-world instruments satisfy time-invariance in practice, the method relaxes a frequent practical barrier to IV use in epidemiology.
  • Testing the estimator on other sampling schemes, such as nested case-control or case-cohort designs, would show how far the bias correction generalizes.

Load-bearing premise

The direct effect of the candidate instrument on the outcome stays constant over time and the selection bias created by retrospective case-control sampling can be modeled and removed.

What would settle it

A simulation or dataset in which the instrument's direct effect on the outcome is allowed to vary over time, yet the iDiD estimator recovers the same numerical value as under the constant-direct-effect assumption.

Figures

Figures reproduced from arXiv: 2607.02327 by Emilie Sbidian, Tat-Thang Vo, Tran Trong Khoi Le.

Figure 1
Figure 1. Figure 1: Directed acyclic graphs. (𝑎): standard instrumental variable; (𝑏): instrumented difference￾in-difference with longitudinal data structure; (𝑐): instrumented difference-in-difference with repeated cross-sectional data structure. In figure 1b, we omit 𝑋 to simplify the graph. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Simulation study: Scenario 1 – Longitudinal data structure. [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Simulation study: Scenario 2 – Repeated cross-sectional data structure. [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illstrative data analysis: selection of case and controls. [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
read the original abstract

Case-control designs are fundamental in epidemiology for the efficient study of rare outcomes. Although instrumental variable (IV) methods have been extended to this setting to address unmeasured confounding, they typically rely on the exclusion restriction assumption, which may be violated when the IV candidates directly affect the outcome through pathways independent of the exposure. In this paper, we propose a novel instrumented difference-in-differences (iDiD) approach tailored to case-control designs. Grounded in structural mean modeling, the proposed method accommodates IV candidates that have time-invariant direct effect on the outcome. When retrospective case-control datasets are collected, the candidate can still be used as a valid instrument on the trend scale when selection bias induced by retrospective sampling is efficiently taken into account. We assess finite-sample performance of this method through extensive simulations, then apply it to evaluate the risk of serious infection of biologic treatments for psoriasis, using French national claim database.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper proposes an instrumented difference-in-differences (iDiD) estimator for retrospective case-control designs, grounded in structural mean models. It extends IV methods to settings where candidate instruments may violate the exclusion restriction via time-invariant direct effects on the outcome, by shifting identification to the trend scale and correcting for selection bias induced by retrospective sampling. The approach is evaluated in simulations and applied to assess infection risks from biologic treatments for psoriasis in French national claims data.

Significance. If the identification and bias-correction arguments hold, the method would meaningfully expand the set of usable instruments in case-control studies of rare outcomes, where unmeasured confounding is common and strict exclusion restrictions are often implausible. The combination of structural mean modeling with explicit retrospective-sampling correction, plus simulation evidence and a real-data application, gives the work practical relevance for epidemiologic causal inference.

major comments (2)
  1. [§3] §3 (identification section): the claim that the instrument remains valid on the trend scale after selection-bias correction rests on the maintained assumption that any direct effect is strictly time-invariant; the manuscript should provide an explicit sensitivity analysis or bounding exercise showing how violations of this assumption affect the iDiD estimand, as this is load-bearing for the central extension beyond standard IV.
  2. [§4] §4 (estimation and asymptotic results): the efficiency claim for the retrospective-sampling correction inside the structural mean model is stated but the precise form of the estimating equations (or the influence function) is not compared to a naive case-control IV estimator; without this comparison it is unclear whether the efficiency gain is first-order or merely finite-sample.
minor comments (3)
  1. [§5] The simulation design in §5 should report the exact parameter values used to generate time-invariant direct effects so that readers can replicate the reported coverage and bias results.
  2. [§6] In the psoriasis application (§6), the chosen candidate instruments and the justification that their direct effects are plausibly time-invariant should be stated more explicitly, ideally with supporting literature.
  3. Notation for the structural mean model parameters (e.g., the trend parameter versus the direct-effect parameter) is introduced without a consolidated table; a small notation table would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and constructive comments. We address each major point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: §3 (identification section): the claim that the instrument remains valid on the trend scale after selection-bias correction rests on the maintained assumption that any direct effect is strictly time-invariant; the manuscript should provide an explicit sensitivity analysis or bounding exercise showing how violations of this assumption affect the iDiD estimand, as this is load-bearing for the central extension beyond standard IV.

    Authors: We agree that the time-invariance of direct effects is a central maintained assumption for identification on the trend scale. While the core results are derived under this assumption, we will add an explicit sensitivity analysis section in the revision, including a bounding exercise that quantifies how violations (time-varying direct effects) affect the iDiD estimand. revision: yes

  2. Referee: §4 (estimation and asymptotic results): the efficiency claim for the retrospective-sampling correction inside the structural mean model is stated but the precise form of the estimating equations (or the influence function) is not compared to a naive case-control IV estimator; without this comparison it is unclear whether the efficiency gain is first-order or merely finite-sample.

    Authors: We agree that an explicit comparison is needed to establish whether the efficiency gain is first-order. In the revised manuscript we will derive the influence function of the proposed estimator, compare it directly to that of a naive case-control IV estimator, and show that the retrospective-sampling correction yields a first-order asymptotic efficiency improvement. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The abstract and description present the iDiD method as an extension grounded in structural mean modeling to accommodate time-invariant direct effects and correct for retrospective sampling bias. No equations, fitted parameters, or derivation steps are shown that reduce by construction to inputs, self-citations, or renamed known results. The approach is described with external validation via simulations and real-data application, making the central claim self-contained against the stated assumptions without load-bearing internal circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no equations, parameters, or modeling details; cannot identify free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5685 in / 1085 out tokens · 33189 ms · 2026-07-03T02:39:04.245223+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

26 extracted references

  1. [1]

    Clinical epidemiology , pages=

    Nordic health registry-based research: a review of health care systems and key registries , author=. Clinical epidemiology , pages=. 2021 , publisher=

    2021

  2. [2]

    European journal of epidemiology , volume=

    Registers of the Swedish total population and their use in medical research , author=. European journal of epidemiology , volume=. 2016 , publisher=

    2016

  3. [3]

    1982 , publisher=

    Case-control studies: design, conduct, analysis , author=. 1982 , publisher=

    1982

  4. [4]

    2008 , publisher=

    Modern epidemiology , author=. 2008 , publisher=

    2008

  5. [5]

    Chest , volume=

    A practical overview of case-control studies in clinical practice , author=. Chest , volume=. 2020 , publisher=

    2020

  6. [6]

    1980 , publisher=

    Statistical methods in cancer research , author=. 1980 , publisher=

    1980

  7. [7]

    Biometrika , volume=

    Logistic disease incidence models and case-control studies , author=. Biometrika , volume=. 1979 , publisher=

    1979

  8. [8]

    Applications to cancer of the lung, breast, and cervix , author=

    A method of estimating comparative rates from clinical data. Applications to cancer of the lung, breast, and cervix , author=. Journal of the National Cancer Institute , volume=. 1951 , publisher=

    1951

  9. [9]

    Statistical Science , volume=

    [Choice as an alternative to control in observational studies]: comment , author=. Statistical Science , volume=. 1999 , publisher=

    1999

  10. [10]

    BJOG: an international journal of obstetrics and gynaecology , volume=

    Hidden biases in observational epidemiology: the case of unmeasured confounding , author=. BJOG: an international journal of obstetrics and gynaecology , volume=

  11. [11]

    The Canadian journal of hospital pharmacy , volume=

    An introduction to health care administrative data , author=. The Canadian journal of hospital pharmacy , volume=

  12. [12]

    American journal of epidemiology , volume=

    The impact of residual and unmeasured confounding in epidemiologic studies: a simulation study , author=. American journal of epidemiology , volume=. 2007 , publisher=

    2007

  13. [13]

    Electronic Journal of Statistics , volume=

    Structural mean models for instrumented difference-in-differences , author=. Electronic Journal of Statistics , volume=. 2024 , publisher=

    2024

  14. [14]

    Biometrics , volume=

    Instrumented difference-in-differences , author=. Biometrics , volume=. 2023 , publisher=

    2023

  15. [15]

    Statistics in Medicine , volume=

    Mendelian randomization analysis of case-control data using structural mean models , author=. Statistics in Medicine , volume=. 2011 , publisher=

    2011

  16. [16]

    Epidemiology , volume=

    Methodological challenges in mendelian randomization , author=. Epidemiology , volume=. 2014 , publisher=

    2014

  17. [17]

    Nature reviews Methods primers , volume=

    Mendelian randomization , author=. Nature reviews Methods primers , volume=. 2022 , publisher=

    2022

  18. [18]

    Epidemiology , volume=

    A practical guide to selection bias in instrumental variable analyses , author=. Epidemiology , volume=. 2019 , publisher=

    2019

  19. [19]

    Epidemiology , volume=

    Instrumental variables: application and limitations , author=. Epidemiology , volume=. 2006 , publisher=

    2006

  20. [20]

    Journal of the American statistical Association , volume=

    Identification of causal effects using instrumental variables , author=. Journal of the American statistical Association , volume=. 1996 , publisher=

    1996

  21. [21]

    Statistics in medicine , volume=

    Instrumental variable methods for causal inference , author=. Statistics in medicine , volume=. 2014 , publisher=

    2014

  22. [22]

    2010 , publisher=

    Causal inference , author=. 2010 , publisher=

    2010

  23. [23]

    Biometrics , volume=

    Polydesigns and causal inference , author=. Biometrics , volume=. 2006 , publisher=

    2006

  24. [24]

    The international journal of biostatistics , volume=

    Designs combining instrumental variables with case-control: estimating principal strata causal effects , author=. The international journal of biostatistics , volume=

  25. [25]

    The international journal of biostatistics , volume=

    Simple optimal weighting of cases and controls in case-control studies , author=. The international journal of biostatistics , volume=

  26. [26]

    Journal of the American Academy of Dermatology , year=

    Infection risk among psoriasis biologic-new users: a cohort study on the French National Health Data System , author=. Journal of the American Academy of Dermatology , year=