Pith. sign in

REVIEW 4 major objections 6 minor 58 references

Leaning Time-Varying Instruments for Identifying Causal Effects in Time-Series Data

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A deep generative method learns time-varying conditional instrumental variables directly from proxy variables, enabling unbiased causal effect estimates under time-varying latent confounders.

desk verdict A useful formalization of time-varying conditional instruments, but the central identifiability claim is assumed, not proved, and the synthetic experiments don't satisfy the paper's own proxy assumption. read the letter →

arxiv 2411.17774 v1 pith:ZA63BX73 submitted 2024-11-26 cs.LG cs.AI

Debo Cheng (1) , Ziqi Xu (2) , Jiuyong Li (1) , Lin Liu (1) , Thuc duy Le (1) , Xudong Guo (1) , Shichao Zhang (3) ((1) University of South Australia (2) RMIT University (3) Guangxi Normal University) This is my paper · ORCID
classification cs.LGcs.AI
keywords causaleffectestimationtime-seriesdatatime-varyinglatentconfoundersconditionalinstrumentalvariablesvariablevariationalautoencoderLSTMinference
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 sets out to show that the average causal effect of a time-varying treatment on a later outcome can be estimated without bias even when unmeasured confounders change over time, by learning a time-varying conditional instrumental variable from observed data. Existing instrumental-variable methods for time series either require an expert to supply the instrument or rely on the stronger assumption of a plain time-varying IV; the proposed TDCIV instead disentangles a latent instrument representation $S_t$ and its conditioning set $Z_t$ from proxy variables using LSTM, VAE, and CVAE components. The central theoretical claim is that, if the learned representations match the assumed full-time causal graph, then $Z_t$ together with the historical record instrumentalises $S_t$ for the effect of $W_t$ on $Y_{t+1}$, so two-stage least squares gives unbiased estimates. A sympathetic reader would care because this would remove a key obstacle to causal effect estimation in healthcare, economics, climate science, and epidemiology when latent confounding is present and no domain-specific instrument is known.

What carries the argument

The central object is the time-varying conditional instrumental variable $S_t$: a variable that behaves as an instrument only after conditioning on a set $Z_t$ and on the historical record. The carrying mechanism is the TDCIV architecture, which uses an LSTM to encode history $\bar{H}_t$, a VAE to disentangle $S_t$ and $Z_t$ from proxy variables in $\bar{X}_t$, and a CVAE to generate the conditioning set; the objective (Equation 10) augments the ELBO with predictors of $W_t$ and $Y_{t+1}$. Theorem 1 is what connects these learned representations to identification: it certifies the three CIV conditions in the full-time DAG, so the extracted $S_t$ and $Z_t$ can be fed to 2SLS for an unbiased effect estimate.

What would settle it

Generate synthetic time series with a known latent time-varying confounder and no proxy variable for $S_t$, so the proxy assumption fails; if TDCIV still removes the confounding bias, the stated premise is not doing the work, and if it fails, the premise is confirmed necessary. Alternatively, train on data generated from a graph that violates Figure 3 by adding a direct edge $S_t \to Y_{t+1}$; the 2SLS estimate should show bias proportional to that edge.

Watch

Extended reading notes

Core claim

The paper claims that time-varying conditional instrumental variables can be learned rather than predefined. In the full-time DAG it proposes, $S_t$ is a latent cause of treatment $W_t$, $Z_t$ is a conditioning set generated from observed covariates, and $Y_{t+1}$ is a sink outcome. Theorem 1 states that if the representations $S_t$ and $Z_t$ shown in Figure 3 can be disentangled and learned from the time-series data, then $Z_t$ together with the historical data $\bar{S}_{t-1}, \bar{W}_{t-1}, \bar{Y}_t, \bar{Z}_{t-1}$ makes $S_t$ a valid conditional instrument for $W_t \to Y_{t+1}$ over time. The proof verifies the three CIV conditions: $S_t$ is associated with $W_t$ given the conditioning set; $S_t$ is independent of $Y_{t+1}$ in the graph with the $W_t \to Y_{t+1}$ edge removed; and $Z_t$ contains no descendants of $Y_{t+1}$. Given that validity, the paper's 2SLS estimator (its Equation 11) recovers the average causal effect $ACE_t(W_t, Y_{t+1})$ at each time step despite time-varying latent confounders.

Load-bearing premise

The load-bearing assumption is that at every time step at least one observed covariate is a proxy for the hidden time-varying conditional instrument, and that the learned representations $S_t$ and $Z_t$ match the assumed full-time causal graph.

Editorial extensions

If this is right

  • A researcher no longer needs to name an instrument ahead of time: the conditioning set and instrument are read from data.
  • Average causal effects can be estimated at each time step with time-varying latent confounders present, a setting where standard sequential-randomisation methods are biased.
  • The method applies beyond simulated data: its climate case study claims to recover known drivers of rainfall as the strongest time-varying causes.
  • Because it handles conditional rather than plain instruments, TDCIV is applicable where the instrument is only valid after adjusting for observed past states.
  • The learned representations feed any CIV-based estimator, so the 2SLS stage could be replaced by a nonlinear estimator without changing the identification argument.

Reading between the lines

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

  • An implication the authors leave implicit is that the practical value of TDCIV hinges on whether the VAE objective identifies $S_t$ and $Z_t$; the paper proves validity of the representations only after they are correctly disentangled, and no identifiability theorem for Equation 10 is provided.
  • A direct testable extension would be to degrade the proxy assumption by adding increasing measurement noise to $\bar{X}_t$; the paper's assumptions predict the bias of the 2SLS estimate should grow, which would isolate the mechanism.
  • The same conditioning-on-the-past logic could be carried over to multiple treatments or survival outcomes, settings where time-varying CIVs are currently not available.
  • If latent confounders also affect $S_t$ directly, the assumed graph in Figure 3 is violated and the d-separation argument breaks; detecting such violations is an open problem.
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

4 major / 6 minor

Summary. The paper proposes TDCIV, a method that combines LSTM and variational autoencoders to learn latent representations of a time-varying conditional instrumental variable (CIV) S_t and its conditioning set Z_t from observed time-series data, and then applies two-stage least squares (2SLS) to estimate the average causal effect of a treatment W_t on an outcome Y_{t+1} in the presence of time-varying latent confounders. The authors define a time-varying CIV in a full-time DAG, state Theorem 1 asserting that if the representations of S_t and Z_t can be disentangled and learned as in their Figure 3, then these representations, together with historical data, instrumentalise S_t relative to W_t -> Y_{t+1}. The paper reports synthetic and real-world climate experiments that appear to show that TDCIV outperforms several causal-effect estimators. However, the central claim that TDCIV actually learns valid time-varying CIV representations without domain knowledge is not established: Theorem 1 is conditional on a disentanglement assumption that is never proved, and the synthetic data generator does not satisfy the paper's own proxy-variable assumption.

Significance. The problem addressed is important: estimating causal effects from time-series data with time-varying latent confounders, while avoiding the need for predefined instrumental variables, would be a valuable contribution to causal inference. The formalization of a time-varying CIV in a full-time DAG (Definition 3) is a useful conceptual step, and the idea of using deep generative models for this task is interesting. If the central claim were rigorously supported, the method would likely be of significant interest to the machine learning and causal inference communities. As the manuscript stands, however, the theoretical result is conditional and the experiments do not test the method under its stated assumptions, so the significance is potential rather than demonstrated.

major comments (4)
  1. [Section III-B (Theorem 1)] Theorem 1 is conditional in a way that leaves the paper's central claim unproved. The theorem states that if the representations of S_t and Z_t 'can be disentangled and learned' as in Figure 3, then they instrumentalise S_t; the proof is only a d-separation check within an assumed graph. No identifiability or disentanglement result is provided for the LSTM-VAE objective in Eq. (10), so the antecedent of the theorem is not established. The abstract and introduction claim that the paper 'theoretically establish[es] the validity of these learned representations,' which is stronger than what is actually shown. This is load-bearing because the 2SLS estimate in Eq. (11) is valid only if the learned S_t satisfies the CIV conditions.
  2. [Section IV-A, Eqs. (12)-(14)] The synthetic data generator violates the proxy-variable assumption of Section III-A. In Eq. (12), X_t is generated only from past values X_{t-i} and W_{t-i} plus noise, with no dependence on the current S_t or Z_t, while S_t and Z_t are generated separately in Eqs. (13)-(14). Thus no observed covariate at time t is a proxy of the latent CIV S_t, contradicting the assumption that at least one proxy variable is available at each step. The experiments therefore do not test TDCIV under the assumptions of Theorem 1, and the low reported errors in Figures 5-6 and Tables II-III cannot independently validate the central claim that the learned S_t is a valid time-varying CIV.
  3. [Section III-A and Definition 3] There is an inconsistency in what S_t is. Definition 3 defines a time-varying CIV as a variable S_t in X_t, while Section III-A states that S_t is a latent factor approximated through measurement errors in the observed covariates X_t, with at least one proxy variable available. These are different data-generating assumptions: if S_t is observed, no latent representation needs to be learned; if S_t is latent and only proxied, the conditions in Definition 3 should be stated for the latent variable and the proxy structure should be made explicit. This ambiguity matters because the validity conditions for 2SLS depend on which variable is the instrument.
  4. [Section III-C, Eq. (10)] The learning objective does not enforce the exclusion restriction or instrument exogeneity for the learned S_t. The term beta * E_q log p(Y_{t+1} | Z_t, H_t) predicts Y_{t+1} without S_t, but nothing prevents S_t from encoding information about Y_{t+1} through H_t or Z_t, and no constraint ensures that S_t is independent of the latent confounders U_t. Without additional identifiability assumptions, infinitely many latent decompositions can fit the ELBO, and the representation extracted as S_t may be correlated with U_t, making it an invalid instrument. The paper's claim that TDCIV learns a valid CIV without domain knowledge is therefore not supported by the objective as stated.
minor comments (6)
  1. [Title] The title contains a typo: 'Leaning' should be 'Learning.'
  2. [Section IV-A, Eq. (14)] The noise term for Z_t is written as epsilon_U, the same symbol used for the latent confounder noise in Eq. (12); this appears to be a typo and should be a distinct noise term.
  3. [Section II-A, Eq. (1)] The equality E[Y_{t+1}(w_t) | W_{t-1}, X_t] = E[Y_{t+1} | w_t, W_{t-1}, X_t] is not true without a no-unmeasured-confounding assumption; as presented it conflates the definition of the ACE with its identification.
  4. [Section III-B, proof of Theorem 1] The proof of condition (i) states that 'S_t and W_t are dependent given any set of covariates,' which is not generally true in a DAG; conditioning on particular sets can create or remove dependence. The specific conditioning set in Definition 3 may be sufficient, but the statement as written is too strong.
  5. [Algorithm 1] The line 'Update parameters of LT DCIV' contains a typo (LT DCIV should be L_TDCIV or TDCIV).
  6. [Section IV-A, evaluation] The paper states that source code will be made publicly accessible upon publication; no code or data repository is provided at submission, which limits reproducibility verification.

Circularity Check

3 steps flagged · score 6.0 of 10

Theorem 1 assumes the very disentanglement it claims to establish, and the Eq. (10) objective fits St to Wt before re-using St as the instrument in Eq. (11); the synthetic generator provides no direct proxy of St to break this circle.

  1. self definitional [Section III-B, Theorem 1]
    "If the representations of St and Zt as shown in the full-time DAG in Figure 3 can be disentangled and learned from time-series data D = { ¯Xt, Wt, Yt}, then Zt along with the historical data { ¯St−1, ¯Wt−1, ¯Yt, ¯Zt−1} instrumentalises St relative to Wt → Yt+1 over time."

    The theorem's conclusion is exactly Definition 3's three CIV conditions. The antecedent already stipulates that St and Zt are learned as in Figure 3, a graph in which St→Wt and St is d-separated from Yt+1 given Zt and history. The proof only checks d-separation in that assumed graph; no identifiability or disentanglement result shows that the Eq. (10) VAE objective yields representations matching Figure 3. Thus the paper's claim that it 'theoretically establish[es] the validity of these learned representations' is conditional on the target property itself: if learned as in the graph, they are valid by that graph's construction.

  2. fitted input called prediction [Section III-C, Eq. (10) and Eq. (11)]
    "To ensure that St captures as much information as possible about the CIV, and that Zt captures more confounding information between St and Yt+1, we add two predictors into our ELBO to predict Wt and Yt+1, respectively... LT DCIV = −M + αEq[log p(Wt | ¯Zt, ¯St, ¯Ht)] + βEq[log p(Yt+1 | ¯Zt, ¯Ht)]... ACE t(Wt, Yt+1) = σSt∗Yt+1∗( ¯Zt, ¯Ht) / σSt∗Wt∗( ¯Zt, ¯Ht)."

    The first added predictor fits St to predict Wt, and Eq. (11) then divides by σ_{St*Wt|Z,H} as the instrument's relevance. Relevance is therefore installed by the training loss rather than discovered. Nothing in Eq. (10) enforces Definition 3(ii), St ⊥ Yt+1 | Zt, history; St can simply encode Wt, including the direct Wt→Yt+1 path. In that case the 2SLS ratio is computed from a learned function of the treatment, not from an independent instrument, so the 'validity' of the CIV is not derived from the data but built into the fitting objective.

1 more flagged steps
  1. other [Section IV-A, Eqs. (12)-(13)]
    "At each time step t, the observed covariates Xt and latent covariates Ut are defined as follows: Xt = 1/p Σ(αi Xt−i + ωi Wt−i) + εX ... The latent time-varying CIV St is generated as follows: St = 1/p Σ(St−i) + εS."

    The synthetic generator used to validate TDCIV has no direct dependence of Xt on St (or on Zt), despite Section III-A assuming 'at least one proxy variable is available for the time-varying CIV at each step' and Section III-B asserting St 'is proxied by one of the observed variables in ¯Xt'. In this data, the only channel through which TDCIV can imbue St with information about the CIV is the log p(Wt | ¯Zt, ¯St, ¯Ht) term of Eq. (10), i.e., through the treatment Wt itself. The reported low absolute errors therefore do not test the proxy-based disentanglement of Theorem 1; they test a regime in which the instrument is fitted from the treatment, completing the same circle.

full rationale

The central theoretical claim is a conditional d-separation lemma. Its proof is internally coherent, but it does not connect the Eq. (10) objective to Figure 3; the abstract's unconditional phrasing, 'we theoretically establish the validity of these learned representations,' overstates what is proved. The more concrete circularity is in the objective: the first added predictor fits St to Wt, and the 2SLS step then re-uses St as the instrument; relevance is fitted, exclusion is neither enforced nor tested, so the estimated ACE is not a first-principles IV estimate. The synthetic design compounds this by not generating any direct proxy of St in Xt, leaving Wt-prediction as the only way to learn St. These are not merely 'not standard consensus' concerns; they are visible in Eqs. (10)-(13). Self-citations to [17], [22], and [55] are present but not load-bearing for these particular steps. The d-separation check itself has independent mathematical content, which keeps the score at 6 rather than 8.

Assumptions & free parameters 3 free parameters · 6 assumptions · 2 invented entities

The central claim rests on standard graphical-causality assumptions (Markov, faithfulness, consistency, LSRA) plus two paper-specific postulates: the existence of proxy variables for the CIV and the ability of the VAE/CVAE objective to recover S_t and Z_t in the assumed graph. The latter is not proven. The method also introduces two latent representations with no independent falsifiable handle. Free parameters include the objective weights alpha/beta, the latent dimensionality, and network sizes.

free parameters (3)
  • alpha, beta loss trade-off weights = not reported
    Eq. (10) balances the ELBO, the treatment prediction term, and the outcome prediction term; no values or sensitivity analysis are reported for these weights.
  • latent dimensions D_St, D_Zt = not reported
    The dimensionalities of the learned CIV and conditioning-set representations are required to implement Eq. (4) but are not specified.
  • LSTM and fully connected hidden units, dropout probability = 128, 128, 0.8
    Table I lists these values but gives no selection or tuning details, and they directly affect representation quality.
assumptions (6)
  • domain assumption The time-series data D is generated by a full-time DAG and satisfies the Markov property and faithfulness (Definitions 1 and 2).
    These standard graphical-causality assumptions are invoked in Section II-A and are needed for d-separation to encode conditional independence.
  • domain assumption Consistency, latent sequential randomization (LSRA), and positivity hold (Assumptions 1, 2, 4).
    These assumptions connect potential outcomes to observed data and define the latent confounding structure; they are stated in Section II-A.
  • domain assumption At least one proxy variable for the latent time-varying CIV St is available in X_t at every time step.
    Stated in Section III-A: the model assumes the CIV can be approximated through measurement errors present in observed covariates. This is not verified in the synthetic experiments because X_t does not directly depend on current S_t.
  • ad hoc to paper The VAE/CVAE objective (Eq. 10) can disentangle and learn St and Zt so they match the causal structure of Figure 3.
    This is the antecedent of Theorem 1 and the paper's main unproven burden. No identifiability or disentanglement guarantee is given for the learned representations.
  • domain assumption Y_{t+1} is a sink node, so Z_t contains no descendants of Y_{t+1}.
    Used in Theorem 1 to satisfy condition (iii) of Definition 3; it is consistent with the assumed graph but is a modeling assumption about the real system.
  • domain assumption The causal effect is linear enough for the 2SLS ratio estimator in Eq. (11) to recover ACE_t.
    Eq. (11) is introduced as applying 'in the linear case'; the generative model in the experiments is linear, but real applications may not be.
invented entities (2)
  • S_t, the learned time-varying CIV representation
    purpose: Serves as the latent instrument in 2SLS to remove confounding from latent time-varying confounders.
    No falsifiable handle outside this paper is provided; validity is assumed by the graph in Figure 3 and only indirectly supported by synthetic effect estimates.
  • Z_t, the learned conditioning-set representation
    purpose: Blocks back-door paths from S_t to Y_{t+1} so S_t becomes a valid conditional instrument.
    A model construct with no independent measurement or prediction that would verify its interpretation as a conditioning set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Leaning Time-Varying Instruments for Identifying Causal Effects in Time-Series Data." pith.science (2026). https://pith.science/paper/ZA63BX73

@misc{pith2026241117774,
  author       = {Pith},
  title        = {Pith review of: Leaning Time-Varying Instruments for Identifying Causal Effects in Time-Series Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZA63BX73}},
  note         = {Machine review of arXiv:2411.17774}
}
read the original abstract

Querying causal effects from time-series data is important across various fields, including healthcare, economics, climate science, and epidemiology. However, this task becomes complex in the existence of time-varying latent confounders, which affect both treatment and outcome variables over time and can introduce bias in causal effect estimation. Traditional instrumental variable (IV) methods are limited in addressing such complexities due to the need for predefined IVs or strong assumptions that do not hold in dynamic settings. To tackle these issues, we develop a novel Time-varying Conditional Instrumental Variables (CIV) for Debiasing causal effect estimation, referred to as TDCIV. TDCIV leverages Long Short-Term Memory (LSTM) and Variational Autoencoder (VAE) models to disentangle and learn the representations of time-varying CIV and its conditioning set from proxy variables without prior knowledge. Under the assumptions of the Markov property and availability of proxy variables, we theoretically establish the validity of these learned representations for addressing the biases from time-varying latent confounders, thus enabling accurate causal effect estimation. Our proposed TDCIV is the first to effectively learn time-varying CIV and its associated conditioning set without relying on domain-specific knowledge.

Figures

Figures reproduced from arXiv: 2411.17774 by the authors.

Figure 1
Figure 1. An example of the Women’s Interagency HIV Study [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) The full-time causal DAG Gfull represents causal relationships over time, where Wt and Yt+1 denote the treatment and outcome, respectively. The time-varying causal effects of Wt on Yt+1 are shown in red. Xt and Ut represent time-varying covariates and latent confounders, influencing both Wt and Yt+1. Blue dashed arrows indicate W1’s effects on subsequent states of all variables, and similar temporal relationship… view at source ↗
Figure 3
Figure 3. We propose a full-time causal DAG Gfull, over time from 1 to t, to disentangle and learn the representations of time-varying CIV and its conditioning set. Our goal is to query ACEt(Wt, Yt+1) over time, as indicated by the red arrow. St and Zt are the learned representations of time-varying CIV and its conditioning set. conditioning on {X¯ t, U¯ t}. However, U¯ t is unmeasured and not included in the time-series data… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: An overview of the TDCIV’s architecture: The LSTM generates the historical data [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Comparison of absolute errors across all methods, presented with their means and standard deviations calculated over [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Absolute error results for all methods, presented with their means and standard deviations over 30 synthetic datasets. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: The time-varying causal effects estimated by our [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: The time-varying causal effects estimated by our [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 51 canonical work pages

  1. [1]

    A. F. Connors, N. V . Dawson et al. , “Outcomes following acute exacerbation of severe chronic obstructive lung disease. the support in- vestigators (study to understand prognoses and preferences for outcomes and risks of treatments),” American Journal of Respiratory and Critical Care Medicine, vol. 154, no. 4, pp. 959–967, 1996

    work page 1996

  2. [2]

    Lifecycle bias in estimates of intergenerational earnings persistence,

    N. D. Grawe, “Lifecycle bias in estimates of intergenerational earnings persistence,” Labour Economics, vol. 13, no. 5, pp. 551–570, 2006

    work page 2006

  3. [3]

    The era-40 re-analysis,

    S. M. Uppala, P. W. K ˚allberg et al., “The era-40 re-analysis,” Quarterly Journal of the Royal Meteorological Society: A Journal of the Atmo- spheric Sciences, Applied Meteorology and Physical Oceanography, vol. 131, no. 612, pp. 2961–3012, 2005

    work page 2005

  4. [4]

    Changes in the duration of european wet and dry spells during the last 60 years,

    O. Zolina, C. Simmer et al., “Changes in the duration of european wet and dry spells during the last 60 years,” Journal of Climate , vol. 26, no. 6, pp. 2022–2047, 2013

    work page 2022

  5. [5]

    Instruments for causal inference: an epidemiologist’s dream?

    M. A. Hern ´an and J. M. Robins, “Instruments for causal inference: an epidemiologist’s dream?” Epidemiology, pp. 360–372, 2006

    work page 2006

  6. [6]

    Causal inference from complex longitudinal data,

    J. M. Robins, “Causal inference from complex longitudinal data,” in Latent variable modeling and applications to causality. Springer, 1997, pp. 69–117

    work page 1997

  7. [7]

    G. W. Imbens and D. B. Rubin, Causal inference in statistics, social, and biomedical sciences . Cambridge University Press, 2015

    work page 2015

  8. [8]

    Data-driven causal effect estimation based on graphical causal modelling: A survey,

    D. Cheng, J. Li, L. Liu, J. Liu, and T. D. Le, “Data-driven causal effect estimation based on graphical causal modelling: A survey,” ACM Computing Surveys, vol. 56, no. 5, pp. 1–37, 2024

    work page 2024

Show all 58 references
  1. [9]

    Marginal structural models to estimate the causal effect of zidovudine on the survival of hiv-positive men,

    M. ´A. Hern ´an, B. Brumback, and J. M. Robins, “Marginal structural models to estimate the causal effect of zidovudine on the survival of hiv-positive men,” Epidemiology, pp. 561–570, 2000

    2000

  2. [10]

    Estimating counterfactual treatment outcomes over time through adversarially balanced representations,

    I. Bica, A. M. Alaa et al., “Estimating counterfactual treatment outcomes over time through adversarially balanced representations,” in Interna- tional Conference on Learning Representations , 2019, pp. 1–28

    2019

  3. [11]

    Time series deconfounder: Estimating treatment effects over time in the presence of hidden confounders,

    I. Bica, A. Alaa et al., “Time series deconfounder: Estimating treatment effects over time in the presence of hidden confounders,” inInternational Conference on Machine Learning . PMLR, 2020, pp. 884–895

    2020

  4. [12]

    Instrumental variable estimation of the marginal structural cox model for time-varying treatments,

    Y . Cui, H. Michael, et al. , “Instrumental variable estimation of the marginal structural cox model for time-varying treatments,” Biometrika, vol. 110, no. 1, pp. 101–118, 2023

    2023

  5. [13]

    Instrumental variable estimation of marginal structural mean models for time-varying treatment,

    H. Michael, Y . Cui et al., “Instrumental variable estimation of marginal structural mean models for time-varying treatment,” Journal of the American Statistical Association , pp. 1–12, 2023

    2023

  6. [14]

    Estimating average causal effects from patient trajectories,

    D. Frauen, T. Hatt et al. , “Estimating average causal effects from patient trajectories,” in Proceedings of the AAAI Conference on Artificial Intelligence, 2023, pp. 7586–7594

    2023

  7. [15]

    Estimating treatment effects from irregular time series observations with hidden confounders,

    D. Cao, J. Enouen et al. , “Estimating treatment effects from irregular time series observations with hidden confounders,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2023, pp. 6897–6905

    2023

  8. [16]

    Causal transformer for estimating counterfactual outcomes,

    V . Melnychuk, D. Frauen, and S. Feuerriegel, “Causal transformer for estimating counterfactual outcomes,” in International Conference on Machine Learning. PMLR, 2022, pp. 15 293–15 329

    2022

  9. [17]

    Instrumental variable estimation for causal inference in longitudinal data with time-dependent latent confounders,

    D. Cheng, Z. Xu et al. , “Instrumental variable estimation for causal inference in longitudinal data with time-dependent latent confounders,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2024, pp. 11 480–11 488

    2024

  10. [18]

    Marginal structural models and causal inference in epidemiology,

    J. M. Robins, M. A. Hernan, and B. Brumback, “Marginal structural models and causal inference in epidemiology,” Epidemiology, pp. 550– 560, 2000

    2000

  11. [19]

    Pearl, Causality

    J. Pearl, Causality. Cambridge university press, 2009

    2009

  12. [20]

    Separators and adjustment sets in causal graphs: Complete criteria and an algorithmic framework,

    B. van der Zander, M. Li ´skiewicz, and J. Textor, “Separators and adjustment sets in causal graphs: Complete criteria and an algorithmic framework,” Artificial Intelligence, vol. 270, pp. 1–40, 2019

    2019

  13. [21]

    Deep IV: A flexible approach for counter- factual prediction,

    J. Hartford, G. Lewis et al., “Deep IV: A flexible approach for counter- factual prediction,” in International Conference on Machine Learning , 2017, pp. 1414–1423

    2017

  14. [22]

    Causal inference with conditional instruments using deep generative models,

    D. Cheng, Z. Xu et al., “Causal inference with conditional instruments using deep generative models,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2023, pp. 7122–7130

    2023

  15. [23]

    Two-stage least squares estimation of average causal effects in models with variable treatment intensity,

    J. D. Angrist and G. W. Imbens, “Two-stage least squares estimation of average causal effects in models with variable treatment intensity,” Journal of the American Statistical Association , vol. 90, no. 430, pp. 431–442, 1995

    1995

  16. [24]

    Generalized instrumental variables,

    C. Brito and J. Pearl, “Generalized instrumental variables,” in Pro- ceedings of the Eighteenth conference on Uncertainty in artificial intelligence, 2002, pp. 85–93

    2002

  17. [25]

    Auto IV: counterfactual prediction via automatic instrumental variable decomposition,

    J. Yuan, A. Wu et al., “Auto IV: counterfactual prediction via automatic instrumental variable decomposition,” ACM Transactions on Knowledge Discovery from Data , vol. 16, no. 4, pp. 74:1–74:20, 2022

    2022

  18. [26]

    Peters, D

    J. Peters, D. Janzing, and B. Sch ¨olkopf, Elements of causal inference: foundations and learning algorithms . The MIT Press, 2017

    2017

  19. [27]

    Long short-term memory,

    S. Hochreiter, “Long short-term memory,” Neural Computation MIT- Press, 1997

    1997

  20. [28]

    Auto-encoding variational bayes,

    D. P. Kingma, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  21. [29]

    Learning structured output representation using deep conditional generative models,

    K. Sohn, X. Yan, and H. Lee, “Learning structured output representation using deep conditional generative models,” in Proceedings of the 28th International Conference on Neural Information Processing Systems- Volume 2, 2015, pp. 3483–3491

    2015

  22. [30]

    S. L. Lauritzen, Graphical models. Clarendon Press, 1996, vol. 17

    1996

  23. [31]

    Marginal structural models versus structural nested mod- els as tools for causal inference,

    J. M. Robins, “Marginal structural models versus structural nested mod- els as tools for causal inference,” in Statistical Models in Epidemiology, the Environment, and Clinical Trials . Springer, 2000, pp. 95–133

    2000

  24. [32]

    Inference on heterogeneous treatment effects in high-dimensional dynamic panels under weak dependence,

    V . Semenova, M. Goldman, V . Chernozhukov, and M. Taddy, “Inference on heterogeneous treatment effects in high-dimensional dynamic panels under weak dependence,” Quantitative Economics , vol. 14, no. 2, pp. 471–510, 2023. [Online]. Available: https: //onlinelibrary.wiley.com/...

    2023 doi

  25. [33]

    Ctp: A causal interpretable model for non-communicable disease progression prediction,

    Z. Sun, W. Zhang et al. , “Ctp: A causal interpretable model for non-communicable disease progression prediction,” arxiv.org/abs/2308.09735, 2023

    2023 arXiv

  26. [34]

    Identifying causal effects using instrumental time series: Nuisance iv and correcting for the past,

    N. Thams, R. Søndergaard et al. , “Identifying causal effects using instrumental time series: Nuisance iv and correcting for the past,” arXiv preprint arXiv:2203.06056, 2022

    2022 arXiv

  27. [35]

    Measurement bias and effect restoration in causal inference,

    M. Kuroki and J. Pearl, “Measurement bias and effect restoration in causal inference,” Biometrika, vol. 101, no. 2, pp. 423–437, 2014

    2014

  28. [36]

    Identifying causal effects with proxy variables of an unmeasured confounder,

    W. Miao, Z. Geng, and E. J. Tchetgen Tchetgen, “Identifying causal effects with proxy variables of an unmeasured confounder,” Biometrika, vol. 105, no. 4, pp. 987–993, 2018

    2018

  29. [37]

    Causal effect inference with deep latent- variable models,

    C. Louizos, U. Shalit et al. , “Causal effect inference with deep latent- variable models,” in Advances in Neural Information Processing Sys- tems, 2017, pp. 6446–6456

    2017

  30. [38]

    An introduction to variational autoencoders,

    D. P. Kingma, M. Welling et al. , “An introduction to variational autoencoders,” Foundations and Trends® in Machine Learning, vol. 12, no. 4, pp. 307–392, 2019

    2019

  31. [39]

    Treatment effect estimation with dis- entangled latent factors,

    W. Zhang, L. Liu, and J. Li, “Treatment effect estimation with dis- entangled latent factors,” in Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI, 2021, pp. 10 923–10 930

    2021

  32. [40]

    The blessings of multiple causes,

    Y . Wang and D. M. Blei, “The blessings of multiple causes,” Journal of the American Statistical Association , vol. 114, no. 528, pp. 1574–1596, 2019

    2019

  33. [41]

    Double/debiased machine learning for treatment and structural parameters,

    V . Chernozhukov, D. Chetverikov et al. , “Double/debiased machine learning for treatment and structural parameters,” The Econometrics Journal, vol. 21, no. 1, pp. C1–C68, 2018

    2018

  34. [42]

    Generalized random forests,

    S. Athey, J. Tibshirani, and S. Wager, “Generalized random forests,” The Annals of Statistics , vol. 47, no. 2, pp. 1148–1178, 2019

    2019

  35. [43]

    Quasi-oracle estimation of heterogeneous treat- ment effects,

    X. Nie and S. Wager, “Quasi-oracle estimation of heterogeneous treat- ment effects,” Biometrika, vol. 108, no. 2, pp. 299–319, 2021

    2021

  36. [44]

    Metalearners for estimating heteroge- neous treatment effects using machine learning,

    S. R. K ¨unzel, J. S. Sekhon et al., “Metalearners for estimating heteroge- neous treatment effects using machine learning,” PNAS, vol. 116, no. 10, pp. 4156–4165, 2019

    2019

  37. [45]

    Marginal structural models to estimate the joint causal effect of nonrandomized treatments,

    M. A. Hern ´an, B. Brumback, and J. M. Robins, “Marginal structural models to estimate the joint causal effect of nonrandomized treatments,” Journal of the American Statistical Association , vol. 96, no. 454, pp. 440–448, 2001

    2001

  38. [46]

    Forecasting treatment responses over time using recurrent marginal structural networks,

    B. Lim, A. Alaa, and M. v. d. Schaar, “Forecasting treatment responses over time using recurrent marginal structural networks,” in Proceedings of the 32nd International Conference on Neural Information Processing Systems, 2018, pp. 7494–7504

    2018

  39. [47]

    EconML: A Python Package for ML- Based Heterogeneous Treatment Effects Estimation,

    K. Battocchi, E. Dillon et al. , “EconML: A Python Package for ML- Based Heterogeneous Treatment Effects Estimation,” pp. 1–6, 2019

    2019

  40. [48]

    Tensorflow: Large-scale machine learning on heterogeneous distributed systems,

    M. Abadi, A. Agarwal et al. , “Tensorflow: Large-scale machine learning on heterogeneous distributed systems,” arXiv preprint arXiv:1603.04467, 2016

    2016 arXiv

  41. [49]

    The ncep/ncar 40-year reanalysis project,

    E. Kalnay, M. Kanamitsu et al. , “The ncep/ncar 40-year reanalysis project,” Bulletin of the American Mseteorological Society, vol. 77, no. 3, pp. 437–471, 1996

    1996

  42. [50]

    A european daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006,

    M. R. Haylock, N. Hofstra et al. , “A european daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006,” Journal of Geophysical Research: Atmospheres, vol. 113, no. D20, 2008

    1950

  43. [51]

    Temperature and cape dependence in observed and modelled convective precipitation over the united states,

    J. T. Seeley and D. M. Romps, “Temperature and cape dependence in observed and modelled convective precipitation over the united states,” Journal of Climate , vol. 33, no. 17, pp. 7305–7322, 2020

    2020

  44. [52]

    Estimating individual treatment effect: generalization bounds and algorithms,

    U. Shalit, F. D. Johansson, and D. A. Sontag, “Estimating individual treatment effect: generalization bounds and algorithms,” in Proceedings of the 34th International Conference on Machine Learning, ICML, 2017, pp. 3076–3085

    2017

  45. [53]

    Disentangled representation for causal mediation analysis,

    Z. Xu, D. Cheng, J. Li, J. Liu, L. Liu, and K. Wang, “Disentangled representation for causal mediation analysis,” inProceedings of the AAAI Conference on Artificial Intelligence , vol. 37, no. 9, 2023, pp. 10 666– 10 674

    2023

  46. [54]

    Causal inference with conditional front-door adjustment and identifiable variational autoencoder,

    Z. Xu, D. Cheng, J. Li, J. Liu, L. Liu, and K. Yu, “Causal inference with conditional front-door adjustment and identifiable variational autoencoder,” in The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024 , 2024. [Onlin...

    2024

  47. [55]

    Conditional instrumental variable regression with representation learning for causal inference,

    D. Cheng, Z. Xu, J. Li, L. Liu, J. Liu, and T. D. Le, “Conditional instrumental variable regression with representation learning for causal inference,” in The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024 , 2024. [Onli...

    2024

  48. [56]

    Instrumental variables: an econometrician’s perspective,

    G. W. Imbens, “Instrumental variables: an econometrician’s perspective,” Statistical Science, vol. 29, no. 3, pp. 323–358, 2014

    2014

  49. [57]

    A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect,

    J. Robins, “A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect,” Mathematical modelling , vol. 7, no. 9-12, pp. 1393–1512, 1986

    1986

  50. [58]

    Instrumental variables estima- tion of exposure effects on a time-to-event endpoint using structural cumulative survival models,

    T. Martinussen, S. Vansteelandt et al. , “Instrumental variables estima- tion of exposure effects on a time-to-event endpoint using structural cumulative survival models,” Biometrics, vol. 73, no. 4, pp. 1140–1149, 2017

    2017

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.