Pith. sign in

REVIEW 2 major objections 4 minor 36 references

Causal interpretation of the sibling comparison and its relation to the cross-over design

T0 review · 2 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read Under two explicit assumptions, the sibling comparison estimates the family-level causal effect; otherwise it estimates a narrower discordant-pair contrast.

desk verdict A genuinely useful mapping of sibling comparisons to cross-over trials, but the headline identification depends on an unstated conditional independence stronger than the paper's home-turf assumption. read the letter →

arxiv 2507.03464 v1 pith:A7CALR2F submitted 2025-07-04 stat.ME

classification stat.ME
keywords siblingcomparisontwindataestimandscross-overdesignaveragecausaltreatmenteffectinterferencestructuraladditivityinverseprobabilityweighting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper asks what the classic sibling comparison actually estimates and gives a precise answer: the matched-pairs estimator computes the exposure-discordant contrast $E[Y_1(1,0)-Y_1(0,1)\mid S=1, X_1\neq X_2]$, not a general exposure effect. It then shows that if two conditions hold, structural additivity and non-differential interference, this contrast equals the family-level effect $T_3=E[Y_1(1,1)-Y_1(0,0)]$. These conditions are exactly what a cross-over trialist calls absence of differential carry-over and absence of trial-by-treatment interaction, making the sibling design an emulated ABBA cross-over trial observed only in its second period. A reader should care because the result converts a vague worry about sibling studies into a pair of assumptions that can be argued from design or probed by weighting.

What carries the argument

The load-bearing object is the emulated within-family ABBA cross-over target trial: families are randomised to exposed-then-unexposed or unexposed-then-exposed arms, and the sibling comparison is treated as the second period of that trial, with the cosibling standing in for the missing first-period observation. The ABBA schematic converts the statistical assumptions into a dictionary, sibling interference is the analogue of carry-over, and deviation from non-differential interference is aliased with differential carry-over $\lambda_1-\lambda_0$, while structural additivity is the analogue of no treatment-by-trial interaction. The formal identity doing the work is the derivation chain in equations (26)-(27), which shows that the matched-pairs estimand equals $E[Y_1(1,0)-Y_1(0,1)\mid S=1, X_1\neq X_2]$ and reduces to $T_3$ under structural additivity and non-differential interference.

What would settle it

Using data with observed family-level covariates, estimate the within-pair exposure contrast from discordant sibling pairs, then re-estimate it after stratifying by a measured proxy of the family outcome level such as parental education or family mean outcome; if the contrast changes systematically with that proxy, structural additivity is violated and the matched estimate cannot equal the family-level effect $T_3$ even when interference is absent.

Watch

Extended reading notes

Core claim

The paper's central claim is an identification result. In the setting where only family-level confounding is present, the between-within estimator of the sibling comparison converges to $E[Y_1(1,0)-Y_1(0,1)\mid S=1, X_1\neq X_2]$, a contrast in which the cosibling is always forced to the opposite exposure within discordant pairs. If structural additivity holds, the conditioning on $X_1\neq X_2$ drops out; if non-differential interference holds, the contrast between $(1,0)$ and $(0,1)$ collapses to the family-level contrast $E[Y_1(1,1)-Y_1(0,0)]$. The proof runs through a DAG with a selection indicator and a family-level confounder $\alpha$, and the final equality is equation (27). The paper also shows that violation of non-differential interference is aliased with differential carry-over in the emulated cross-over design, so the two assumptions are the same ones a cross-over trialist needs for non-differential carry-over and no treatment-by-trial interaction.

Load-bearing premise

The load-bearing premise is that all unmeasured confounding is family-constant, so that a sibling's exposure is independent of sibling-specific unobserved causes of the outcome once the family-level component $\alpha$ is fixed; the paper itself flags this home-turf assumption in the final paragraph of the Discussion, and if it fails, the matched analysis does not identify even the discordant-pair contrast.

Editorial extensions

If this is right

  • Discordant-pair estimates from sibling comparisons should not be reported as population- or family-level average effects unless structural additivity and non-differential interference are defended; the default target is the narrower discordant contrast.
  • The between-within random-intercept model and the fixed-effects matched-pairs analysis share the same limiting target, so the choice between them affects estimation and estimation of $\alpha$, not which effect is being identified.
  • When structural additivity or non-differential interference fails, the weighting estimator can still target conditional family-level effects such as $E[Y_1(1,1)-Y_1(0,0)\mid S=1]$, provided a correct model for exposure given $\alpha$ and a good estimate of $\alpha$ are available.
  • Disagreement between the weighted and between-within estimates can be read as evidence against structural additivity or non-differential interference, giving a sensitivity check rather than a replacement estimator.
  • With a population-level reference sample containing outcome and exposure data, a second weighting stage can remove selection into the source population and estimate unconditional targets, under the assumption that selection depends on confounders only through $\alpha$.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: the cross-over dictionary suggests a diagnostic not discussed in the paper, comparing period-specific within-family contrasts in a sibling panel with temporally ordered exposures could estimate $\lambda_1-\lambda_0$ directly and thereby test non-differential interference rather than assuming it.
  • Editorial extension: because the weighting method needs an estimate of the unobserved family confounder $\alpha$, it is likely to be most trustworthy as a sensitivity analysis for the matched estimate, despite its theoretical appeal as an estimator of other targets.
  • Editorial extension: the home-turf premise of no sibling-specific confounding, which the paper's final paragraph explicitly flags, means the whole equivalence collapses if unmeasured causes of exposure and outcome vary within a family; a natural extension would be a formal bias analysis for such non-shared confounding operating alongside the selection-by-$\alpha$ mechanism.
  • Editorial extension: the result also suggests that twin registries reporting discordant-pair results should report the weighted family-level contrast alongside them, since the gap between the two is a direct operational measure of how much the discordant target differs from the family-level target.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper gives a causal-target interpretation of sibling comparison designs by drawing an analogy with the ABBA cross-over trial. It defines three targets (T1, T2, T3), introduces weak and strong versions of no-interference and a non-differential interference (NDI) condition, and shows that the matched-pairs estimator targets a discordant-pair contrast that, under structural additivity (SA) and NDI, equals the family-level effect T3. The authors then propose inverse-probability-type weights to estimate T3 without relying on SA and NDI, and they evaluate the weighting procedure in simulations and in a twin-data example on familial conflict and quality of life. The core algebraic derivations in Sections 3-5 are internally consistent and the cross-over analogy is presented in detail.

Significance. If the assumptions are granted, the paper makes a useful conceptual contribution: it translates the two obstacles to interpreting sibling comparisons (selection into discordant pairs and interference from the cosibling's exposure) into the language of carry-over and treatment-by-trial interaction in cross-over trials. The explicit step-by-step derivation in equation (26) and the clean definition of NDI in Section 3 are valuable pedagogical and technical contributions. The paper is also honest about the practical limitations of the weighting proposal. However, the central identification result is less general than the paper's stated home-turf assumption suggests, and the weighting method's usefulness as a sensitivity check is not fully supported by the simulation evidence. These issues are fixable within the manuscript's scope.

major comments (2)
  1. [§5, Eq. (26), step (∗), and §5.1, Eq. (30)] The identification of the matched estimator as T3 under SA and NDI depends on the step labelled (∗), where f_{U|α,X1,X2,S} is replaced by f_{U|α,S}. This is the conditional independence U⊥(X1,X2)|α,S=1. The DAG in Figure 3 and the surrounding text allow U to affect X1 and X2 directly, in which case this independence does not follow from the graph. The condition is stronger than the 'no non-constant confounding' caveat in the final Discussion paragraph: a family-level variable that affects both siblings' exposures and modifies the exposure effect is constant within families but violates (∗). A concrete counterexample is Y_j = α + β_W X_j + β_D X_j U + ε_j with symmetric exposure probabilities depending on both U and α and with U independent of α. This model satisfies SA and NDI, yet the matched estimand is β_W + β_D E[U|X1≠X2], not T3 = β_W + β_D E[U]. The same independence is used in the weighting derivation in equation (30), so the weighting proposal does not circumvent this assumption. The paper should state (∗) as an explicit assumption, distinguish it from no non-constant confounding, and provide a sensitivity analysis over the strength of direct U→X1,X2 effects.
  2. [§6, Table 1] The simulation results do not yet support the weighting method as a dependable sensitivity check. When the weights are evaluated at an estimated α, the weighted estimator has bias -5.31 in Scenario 1, -3.32 in Scenario 2, and -2.61 in Scenario 3, and in Scenarios 1 and 2 the bias is larger in magnitude than the bias of the between-within model it is meant to check. The text in Section 6 acknowledges the larger bias but still proposes that a discrepancy between the two procedures can be taken to 'indicate a violation of the (SA) and (NDI) assumptions.' This inference is not supported by the simulations: under Scenarios 1-3 both estimators are biased, so the discrepancy may reflect failure to estimate α well under a misspecified model rather than a violation of SA or NDI. Because the validity-check use is one of the two advertised uses of the weighting procedure, the paper should either demonstrate that the sign or magnitude of the discrepancy tracks SA/NDI violations independently of α-estimation error, or restrict the claim to settings where α can be estimated reliably.
minor comments (4)
  1. [§6, first paragraph] The sentence 'The two methods are viewed as targeting the conditional version of T3 given selection, i.e. E[Y1(1,0) − Y1(0,1)|S=1, X1=x, X2=1−x]' is inconsistent with the definition of T3 in Section 3 and with the simulation targets in Table 1; the displayed expression is the discordant-pair contrast, not T3. Please correct the wording.
  2. [§5.1, Eq. (29)] The weight definition contains a typographical error: 'Yij w(αi, x1.x2)' should presumably be 'Yij w(αi, x1, x2)' with a comma.
  3. [Throughout] The manuscript needs proofreading for typos and repeated words, including 'Correspondance' (footnote), 'conventional analysis conventional analysis' (§4), 'abscence' (§5), 'preceeding' and 'litterature' (Discussion), 'questionaire' and 'presumeably' (§7), 'eights observations' (§7), and 'modellering' (Discussion).
  4. [Figure 3 and §5] Given that the independence (∗) is load-bearing, the DAG in Figure 3 should make explicit whether U has direct edges to X1 and X2. As drawn and described, the graph is ambiguous on this point, which is precisely the source of the stronger-than-stated assumption discussed in the major comments.

Circularity Check

1 steps flagged · score 6.0 of 10

The proposed weighting check for (SA)/(NDI) is partially circular because it estimates the nuisance confounder alpha from the very between-within model whose assumptions it is meant to assess.

  1. fitted input called prediction [Section 6, paragraph beginning 'Of course this raises the question'; see also Eq. (31) and Table 1]
    "First, it may be used to assess the assumptions of (SA) and (NDI) for the BW model matched analysis. The argument is one of contradiction: If one believes the BW model to “hold”, so that (SA) and (NDI) is satisfied, then one would expect it to yield good estimates of α, and then the estimates from the BW model and weighted analysis are expected to agree. Thus, one may take discrepant estimates between the two procedures to indiate a violation of the (SA) and (NDI) assumptions."

    The weights require alpha, which Section 5.1 proposes to estimate from the between-within model (Eq. 31). The paper's own simulations show that when (SA)/(NDI) fail, the BW model is biased and the estimated-alpha weighting is also badly biased (Table 1: bias -5.31, -3.32, -2.61 in Scenarios 1-3). Therefore the 'contradiction check' uses a weight model whose construction already presupposes the model being tested: if the model fails, the alpha estimate is unreliable, the weights are misspecified, and disagreement cannot be attributed to the assumption violation. Agreement is built in when the model holds, and disagreement is uninterpretable when it does not; the check reduces to the fitted model.

full rationale

The formal derivation of the matched-pairs target and its equality to T3 under (SA)+(NDI) (Eqs. 26-27) is a genuine derivation from explicit causal assumptions rather than a circular one; it reproduces Petersen and Lange (2020) with explicit conditioning on S=1 and does not smuggle the conclusion into the premises. The cross-over analogy is expository, not load-bearing. The final caveat that the design works only on its 'home-turf' of no non-constant confounding is a stated structural limitation, not a hidden circular step; although step (*) (U independent of (X1,X2) given alpha, S=1) is stronger than the paper's informal phrasing, it is a stated assumption rather than a derived result. The one genuine circularity is the proposed use of the weighting procedure as a validity check for the matched analysis: the weights require alpha estimated from the between-within model, whose validity and unbiasedness are exactly what the check is meant to test. The paper's own simulations show the estimated-alpha weighting is badly biased whenever SA or NDI fails, so the contradiction check cannot distinguish assumption violation from weight misspecification; the check's 'prediction' reduces to the fitted model. This is a partial circularity in a secondary but advertised contribution; the central causal identification itself remains independent.

Assumptions & free parameters 2 free parameters · 11 assumptions · 0 invented entities

The central identification rests on a battery of causal and distributional assumptions, most notably the no-non-constant-confounding condition and the DAG independence (X1,X2) ⊥ U | α, S=1. The weighting method additionally requires a correct weight model and valid BLUP estimates of α, both of which are shown to be fragile in the simulation study. No new physical or conceptual entities are introduced; α is a standard family-level random intercept.

free parameters (2)
  • Weight model coefficients (logistic intercepts and slopes on α) for exposure concordance = estimated from data in the example and simulations
    The weighting method's unbiasedness relies on a correctly specified (or at least containing) model for P(X1=x1,X2=x2|α,S=1); these coefficients are fitted and their estimation contributes to the bias observed in Table 1 when α is estimated.
  • BLUP estimates of family confounder α (bα_i) = estimated from between-within model
    The weighting procedure evaluates weights at an estimated α. Simulation results show this substitution introduces large bias (e.g., -5.31 in Scenario 1), so the method's practical performance depends critically on this fitted quantity.
assumptions (11)
  • domain assumption Symmetry of sibling distribution (S1): (X1,X2,Y1,Y2) ∼ (X2,X1,Y2,Y1)
    Used to equate expectations across siblings; invoked in Section 2 and throughout the derivations.
  • domain assumption Conditional exchangeability (fC1) with symmetric confounding: Yj(x1,x2) ⊥ (X1,X2) | U, α, S=1 and P(α∈·|X1=x1,X2=x2,S=1)=P(α∈·|X1=x2,X2=x1,S=1)
    Standard causal assumption needed for identification; stated in Section 5.
  • domain assumption Consistency (C2): Yj(x1,x2)=Yj on {X1=x1,X2=x2}
    Connects counterfactuals to observed outcomes; invoked in equation (26).
  • domain assumption Representativity (R): E[Y1(1,0)-Y1(0,1)|S=1,α]=E[Y1(1,0)-Y1(0,1)|α]
    Transportability assumption needed to generalize from the study population to the target population; equation (24).
  • domain assumption Structural additivity (SA) holds in mean: the discordant contrast is independent of α
    Required for the matched analysis to target a population effect rather than a discordant-subpopulation effect; equation (25) and used in equation (27).
  • domain assumption Non-differential interference (NDI): E[Y1(1,1)-Y1(1,0)] = E[Y1(0,0)-Y1(0,1)]
    Required to equate the discordant contrast to the family-level target T3; used in equation (27) and in the cross-over mapping.
  • domain assumption No non-constant confounding (all confounding is family-constant)
    Stated in the final paragraph of the Discussion as the absolute home-turf of the sibling comparison; if false, the entire causal interpretation collapses.
  • domain assumption Conditional independence (X1,X2) ⊥ U | α, S=1 (step (∗) in equation (26))
    Needed to drop U from the conditioning set in the derivation; this is a structural assumption that α captures all exposure-relevant unmeasured confounding.
  • domain assumption Weight model contains the true exposure model
    Required for the weighting estimator to be unbiased; the paper notes in Section 6 that the procedure depends on a correct weight model.
  • domain assumption Between-within model for α with Gaussian random effects, so BLUP is a valid plug-in
    The paper notes Gaussian assumptions are only used to justify equality between conditional and GLS estimators; the BLUP estimates are used as plug-in values for α in weighting.
  • domain assumption Independence of families (S2)
    Observations on different families are independent; stated in Section 2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal interpretation of the sibling comparison and its relation to the cross-over design." pith.science (2026). https://pith.science/paper/A7CALR2F

@misc{pith2026250703464,
  author       = {Pith},
  title        = {Pith review of: Causal interpretation of the sibling comparison and its relation to the cross-over design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A7CALR2F}},
  note         = {Machine review of arXiv:2507.03464}
}
read the original abstract

The intuitive motivation for employing a sibling comparison design is to adjust for confounding that is constant within families. Such confounding can be caused by variables that otherwise might prove difficult to measure, for example factors relating to genetics, environment, and upbringing. Recent methodological investigations have shown that despite its intuitive appeal, the conventionally employed analysis does not relate to a well-defined causal target, even in the case of constant confounding. A main challenge is that the analysis will target the subpopulation of exposure discordant pairs. In the presence of an effect of the cosibling's exposure on the sibling's outcome, there is a second challenge, namely that the effect corresponds to an intervention that always exposes the cosibling to the opposite exposure from the sibling. We characterise the sibling comparison in terms of the cross-over design. Estimands of interest are discussed before using this characterisation to establish more natural conditions for targeting an appropriate causal parameter. We cast the above-mentioned challenges of the sibling comparison in terms of those facing the cross-over trialist: in order to target an appropriate estimand one must be able to argue the absence of a certain type of carry-over effect as well as absence of trial-by-treatment interaction, thus establishing that the former study design emulates the latter warts and all. We explore weighting to counter the effects of such interactions and to target other estimands. The weights rely on estimates of the unobserved confounding structure. Through simulations and an example analysis, we illustrate its potential usefulness to assess the validity of the assumptions of the matched analysis. We briefly discuss an extension of the weighting procedure to remove selection bias based on data from a population-level reference sample.

Figures

Figures reproduced from arXiv: 2507.03464 by the authors.

Figure 1
Figure 1. DAG representing the assumed relationship between the outcomes [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Schematic of a within-family ABBA cross-over design, as described in the text. The [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. DAG representing a potential expansion of the setup in Figure [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Plot of estimates for T3 in Scenario 4. The horisontal axis is the estimates from the BW model, while the vertical axis shows estimates from the weighting procedure evaluating the estimated weight at estimated α. Xj = 1{F Cj<−15} to be the exposure, so that a sibling i…
Figure 5
Figure 5. Figure 5: Plot of the weights ˜w(X1, X2, α) = (1 − X1) · (1 − X2) · w(0, 0, α) + X1 · X2 · w(1, 1, α) with the weight function w as defined in equation (29). The weight is evaluated at the estimated α and coloured by the overall family exposure status. Points have been jittered …
Figure 6
Figure 6. Figure 6: Classic matching: DAG representing the assumed relationship between the outcomes [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: Slight simplification of the DAG in Figure [PITH_FULL_IMAGE:figures/full_fig_p026_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 31 canonical work pages

  1. [1]

    Brumback, Li Li, and Zhuangyu Cai

    Babette A. Brumback, Li Li, and Zhuangyu Cai. On the use of between–within models to adjust for confounding due to unmeasured cluster-level covariates. Communications in Statistics - Simulation and Computation, 46 0 (5): 0 3841--3854, 2017. doi:10.1080/03610918.2015.1024858

  2. [2]

    Regression models for twin studies: A critical review

    John B Carlin, Lyle C Gurrin, Jonathan AC Sterne, Ruth Morley, and Terry Dwyer. Regression models for twin studies: A critical review. International Journal of Epidemiology, 34 0 (5): 0 1089--1099, October 2005. doi:10.1093/ije/dyi153

  3. [3]

    Carroll, David Ruppert, Leonard A

    Raymond J. Carroll, David Ruppert, Leonard A. Stefanski, and Ciprian M. Crainiceanu. Measurement Error in Nonlinear Models : A Modern Perspective , Second Edition . Chapman and Hall/CRC , New York, 2 edition, 2006. ISBN 978-0-429-13963-5. doi:10.1201/9781420010138

  4. [4]

    Tattoo ink exposure is associated with lymphoma and skin cancers -- a Danish study of twins

    Signe Bedsted Clemmensen, Jonas Mengel-From , Jaakko Kaprio, Henrik Frederiksen, and Jacob von Bornemann Hjelmborg . Tattoo ink exposure is associated with lymphoma and skin cancers -- a Danish study of twins. BMC Public Health, 25 0 (1), January 2025. doi:10.1186/s12889-025-21413-3

  5. [5]

    Mixed Models : Theory and Applications with R

    Eugene Demidenko. Mixed Models : Theory and Applications with R . John Wiley & Sons, Incorporated, Somerset, United States, 2013. ISBN 978-1-118-59306-6

  6. [6]

    Improving transportability of randomized controlled trial inference using robust prediction methods

    Michael R Elliott, Orlagh Carroll, Richard Grieve, and James Carpenter. Improving transportability of randomized controlled trial inference using robust prediction methods. Statistical Methods in Medical Research, 32 0 (12): 0 2365--2385, December 2023. doi:10.1177/09622802231210944

  7. [7]

    ICH E9 ( R1 ) addendum on estimands and sensitivity analysis in clinical trials to the guideline on statistical principles for clinical trials

    EMA/CHMP/ICH. ICH E9 ( R1 ) addendum on estimands and sensitivity analysis in clinical trials to the guideline on statistical principles for clinical trials. Technical report, European Medicines Agency, 2020

  8. [8]

    P. R. Freeman. The performance of the two-stage analysis of two-treatment, two-period crossover trials. Statistics in Medicine, 8 0 (12): 0 1421--1432, 1989. doi:10.1002/sim.4780081202

Show all 36 references
  1. [9]

    Sibling Comparison Designs : Bias From Non - Shared Confounders and Measurement Error

    Thomas Frisell, Sara Öberg, Ralf Kuja-Halkola, and Arvid Sjölander. Sibling Comparison Designs : Bias From Non - Shared Confounders and Measurement Error . Epidemiology, 23 0 (5): 0 713--720, September 2012. doi:10.1097/EDE.0b013e31825fa230

  2. [10]

    Sander Greenland, Judea Pearl, and James M. Robins. Causal Diagrams for Epidemiologic Research . Epidemiology, 10 0 (1): 0 37--48, January 1999

  3. [11]

    Causal Inference : What If

    Miguel A Hernan and James M Robins. Causal Inference : What If . Chapman & Hall/CRC, Boca Raton, 2020

  4. [12]

    Design and analysis of cross-over trials

    Byron Jones and Michael G Kenward. Design and analysis of cross-over trials. Chapman and Hall/CRC, 2003

  5. [13]

    Kalow, B

    W. Kalow, B. K. Tang, and L. Endrenyi. Hypothesis: comparisons of inter- and intra-individual variations can substitute for twin studies in drug research. Pharmacogenetics, 8 0 (4): 0 283--289, August 1998. doi:10.1097/00008571-199808000-00001

  6. [14]

    Laird, Jonathan Skinner, and Michael Kenward

    Nan M. Laird, Jonathan Skinner, and Michael Kenward. An analysis of two-period crossover designs with carry-over effects. Statistics in Medicine, 11: 0 1967--1979, 1992. doi:10.1002/sim.4780111415

  7. [15]

    Using principal stratification in analysis of clinical trials

    Ilya Lipkovich, Bohdana Ratitch, Yongming Qu, Xiang Zhang, Mingyang Shan, and Craig Mallinckrodt. Using principal stratification in analysis of clinical trials. Statistics in Medicine, 41 0 (19): 0 3837--3877, 2022. doi:10.1002/sim.9439

  8. [16]

    Can we generalise from findings in twins? Paediatric and Perinatal Epidemiology, 19 0 (s1): 0 54--59, 2005

    Ruth Morley. Can we generalise from findings in twins? Paediatric and Perinatal Epidemiology, 19 0 (s1): 0 54--59, 2005. doi:10.1111/j.1365-3016.2005.00610.x

  9. [17]

    o ren M \

    Lorelei A. Mucci, Jacob B. Hjelmborg, Jennifer R. Harris, Kamila Czene, David J. Havelick, Thomas Scheike, Rebecca E. Graff, Klaus Holst, S \"o ren M \"o ller, Robert H. Unger, Christina McIntosh, Elizabeth Nuttall, Ingunn Brandt, Kathryn L. Penney, Mikael Hartman, Peter Kraft...

  10. [18]

    J. M. Neuhaus and J. D. Kalbfleisch. Between- and within-cluster covariate effects in the analysis of clustered data. Biometrics, 54 0 (2): 0 638--645, 1998

  11. [19]

    Neuhaus and Charles E

    John M. Neuhaus and Charles E. McCulloch. Separating between- and within- Cluster Covariate Effects by Using Conditional and Partitioning Methods . Journal of the Royal Statistical Society. Series B (Statistical Methodology), 68 0 (5): 0 859--872, 2006

  12. [20]

    What is the causal interpretation of sibling comparison designs? Epidemiology, 31 0 (1): 0 75--81, 2020

    Anne Helby Petersen and Theis Lange. What is the causal interpretation of sibling comparison designs? Epidemiology, 31 0 (1): 0 75--81, 2020

  13. [21]

    Representation of the people? The Lancet, 331 0 (8598): 0 1346, June 1988

    Richard Peto. Representation of the people? The Lancet, 331 0 (8598): 0 1346, June 1988. doi:10.1016/S0140-6736(88)92167-8

  14. [22]

    Stig Hebbelstrup Rye Rasmussen, Steven Ludeke, and Jacob V. B. Hjelmborg. A Major Limitation of the Direction of Causation Model : Non-Shared Environmental Confounding . Twin Research and Human Genetics, 22 0 (1): 0 14--26, February 2019. doi:10.1017/thg.2018.67

  15. [23]

    Review of methods for handling confounding by cluster and informative cluster size in clustered data

    Shaun Seaman, Menelaos Pavlou, and Andrew Copas. Review of methods for handling confounding by cluster and informative cluster size in clustered data. Statistics in Medicine, 33 0 (30): 0 5371--5387, 2014. doi:10.1002/sim.6277

  16. [24]

    Review of inverse probability weighting for dealing with missing data

    Shaun R Seaman and Ian R White. Review of inverse probability weighting for dealing with missing data. Statistical Methods in Medical Research, 22 0 (3): 0 278--295, 2011. doi:10.1177/0962280210395740

  17. [25]

    S. J. Senn and H. Hildebrand. Crossover trials, degrees of freedom, the carryover problem and its dual. Statistics in Medicine, 10 0 (9): 0 1361--1374, 1991. doi:10.1002/sim.4780100905

  18. [26]

    S. J. Senn, P. Auclair, and Samuel Johnson. The graphical representation of clinical trials with particular reference to measurements over time. Statistics in Medicine, 9 0 (11): 0 1287--1302, 1990. doi:10.1002/sim.4780091108

  19. [27]

    Statistical issues in bioequivalance

    Stephen Senn. Statistical issues in bioequivalance. Statistics in Medicine, 20 0 (17-18): 0 2785--2799, 2001. doi:10.1002/sim.743

  20. [28]

    A cautionary note on the recently proposed ICE Falcon method

    Arvid Sj \"o lander and Thomas Frisell. A cautionary note on the recently proposed ICE Falcon method. International Journal of Epidemiology, 53 0 (5), September 2024. doi:10.1093/ije/dyae131

  21. [29]

    o lander, Thomas Frisell, and Sara \

    Arvid Sj \"o lander, Thomas Frisell, and Sara \"O berg. Sibling Comparison Studies . Annual Review of Statistics and Its Application, 9 0 (1): 0 71--94, March 2022 a . doi:10.1146/annurev-statistics-040120-024521

  22. [30]

    o lander, Sara \

    Arvid Sj \"o lander, Sara \"O berg, and Thomas Frisell. Generalizability and effect measure modification in sibling comparison studies. European Journal of Epidemiology, 37 0 (5): 0 461--476, May 2022 b . doi:10.1007/s10654-022-00844-x

  23. [31]

    Confounders, Mediators , or Colliders : What Types of Shared Covariates Does a Sibling Comparison Design Control For ? Epidemiology, 28 0 (4): 0 540--547, July 2017

    Arvid Sjölander and Johan Zetterqvist. Confounders, Mediators , or Colliders : What Types of Shared Covariates Does a Sibling Comparison Design Control For ? Epidemiology, 28 0 (4): 0 540--547, July 2017. doi:10.1097/EDE.0000000000000649

  24. [32]

    Causal Interpretation of Between - Within Models for Twin Research

    Arvid Sjölander, Thomas Frisell, and Sara Öberg. Causal Interpretation of Between - Within Models for Twin Research . Epidemiologic Methods, 1 0 (1), January 2012. doi:10.1515/2161-962X.1015

  25. [33]

    Carryover Effects in Sibling Comparison Designs

    Arvid Sjölander, Thomas Frisell, Ralf Kuja-Halkola, Sara Öberg, and Johan Zetterqvist. Carryover Effects in Sibling Comparison Designs . Epidemiology, 27 0 (6): 0 852--858, November 2016. doi:10.1097/EDE.0000000000000541

  26. [34]

    Harris, Kamila Czene, Lorelei Mucci, Hans-Olov Adami, Kaare Christensen, Jacob Hjelmborg, Niels V

    Axel Skytthe, Jennifer R. Harris, Kamila Czene, Lorelei Mucci, Hans-Olov Adami, Kaare Christensen, Jacob Hjelmborg, Niels V. Holm, Thomas S. Nilsen, Jaakko Kaprio, and Eero Pukkala. Cancer Incidence and Mortality in 260,000 Nordic Twins With 30,000 Prospective Cancers . Twin R...

  27. [35]

    Boomsma, Irene Rebollo-Mesa , James J

    Niels van der Aa, Dorret I. Boomsma, Irene Rebollo-Mesa , James J. Hudziak, and Meike Bartels. Moderation of Genetic Factors by Parental Divorce in Adolescents ' Evaluations of Family Functioning and Subjective Wellbeing . Twin Research and Human Genetics, 13 0 (2): 0 143--162...

  28. [36]

    VanderWeele and Miguel A

    Tyler J. VanderWeele and Miguel A. Hern \'a n. Causal Inference Under Multiple Versions of Treatment . Journal of causal inference, 1 0 (1): 0 1--20, May 2013. doi:10.1515/jci-2012-0002

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.