Open research questions in Advanced Causal Inference Techniques

What the literature leaves open

• Further research can be done to explore the extension to multivariate covariates. More simulation studies can be conducted to consider different models for the potential outcomes.

  CTE2: Conditional Tail Expectation Treatment Effect · 2026 · DOI

• The lack of a method to evaluate the effect of treatment for extreme subpopulations in a way that is more interpretable than the pointwise CATE. The need for a novel estimator that can handle extreme subpopulations.

  CTE2: Conditional Tail Expectation Treatment Effect · 2026 · DOI

• The gap in the literature is the lack of work on multivalued treatments with discrete-valued instruments. The paper identifies the need for a framework to analyze the effects of multivalued treatments. The existing work on multivalued treatments under selection on observables has limitations.

  Treatment effects with targeting instruments · 2026 · DOI

• Nonseparable or structural models with sufficiently flexible specifications of site-specific heterogeneity may be another fruitful approach to this problem

  Transfer estimates for causal effects across heterogeneous sites · 2026 · DOI

• The lack of a method to adapt experimental estimates to new locations. The inability to account for site-specific heterogeneity in existing approaches. The need for a nonparametric approach to predict causal responses using baseline data from target sites.

  Transfer estimates for causal effects across heterogeneous sites · 2026 · DOI

• The paper identifies the challenge of estimating treatment effects when facing limited overlap. The paper highlights the challenge of constructing robust inference methods, particularly when the overlap condition is violated. The paper discusses the challenge of generalizing the bounds to higher-order bounds, which requires careful consideration of the asymptotic analysis.

  Bounding treatment effects by pooling limited information across observations · 2026 · DOI

• The paper identifies a gap in the literature, where existing approaches may not be robust in challenging situations, such as when the conditioning variables take on a large number of different values. The paper highlights the need for more robust inference methods, particularly when the overlap condition is violated.

  Bounding treatment effects by pooling limited information across observations · 2026 · DOI

• The paper does not provide a comprehensive evaluation of the proposed methods in small samples - The paper notes that it is unclear whether the strata size is sufficiently large in some scenarios - The paper recommends the application of different methods when the experimental design involves a matched-tuples design with only one or two observations per treatment arm

  Inference for two-stage experiments under covariate-adaptive randomization · 2026 · DOI

• Commonly used inference methods can lead to either conservative or invalid inference when ignoring covariate information. Ignoring covariate information in the design stage can result in efficiency loss.

  Inference for two-stage experiments under covariate-adaptive randomization · 2026 · DOI

• JMs require advanced statistical expertise, - existing software is not applicable in all situations, - limited number of variables, - types of associations, - residual bias and low coverage rates for the 95% confidence intervals in cases of strong association or large measurement error

  Including an infrequently measured time-varying error-prone covariate in survival analyses: a simulation-based comparison of methods · 2026 · DOI

• Investigating the performance of simpler approaches, - Exploring the application of MI and JM techniques in various contexts

  Including an infrequently measured time-varying error-prone covariate in survival analyses: a simulation-based comparison of methods · 2026 · DOI

• The gap between the validity and relevance of target trials. The challenge of ensuring that the assumptions on which the analysis depends are reasonable and assessed to the greatest extent possible. The need to balance validity and relevance in target trials.

  What phases of the drug development framework can epidemiological studies emulate? · 2026 · DOI

• The gap in current research is the lack of understanding of the applicability of the two-arm approximation to binary outcomes. The paper identifies the need for a comprehensive evaluation of the two-arm approximation for binary outcomes. The gap also includes the lack of interactive tools for sizing 2k factorial optimization trials.

  Power Calculation in 2k Factorial Optimization Trials: An Interactive Web Application, with Investigation of Robustness for Binary Outcome Variables · 2026 · DOI

• Researchers should consider replacing the functional form assumptions in two cases, - More flexible tests are useful when researchers suspect a non-linear association between the NCO and the IV or between the NCI and the outcome

  Negative Control Falsification Tests for Instrumental Variable Designs · 2026 · DOI

• The identification assumptions in instrumental variable designs cannot be directly tested. Conventional applications of falsification tests may flag problems even in valid IV designs.

  Negative Control Falsification Tests for Instrumental Variable Designs · 2026 · DOI

• The existing literature on sensitivity analyses does not fully address the concern that omitted variables can flip coefficient signs. Many methods use breakdown points as quantitative measures of robustness, but these points can be misleading. There is a need for a modified measure of robustness to address this concern.

  The Effect of Omitted Variables on the Sign of Regression Coefficients · 2026 · DOI

• The specific mechanisms involved in the social contagion of DUD from one sibling to another are not fully understood. Prior work has not fully examined the causal impact of older siblings' academic achievement on younger siblings' risk for drug use disorder.

  The causal impact of older siblings’ academic achievement on younger siblings’ risk for drug use disorder: instrumental variable and propensity score analyses · 2026 · DOI

• The unconstrained PPO model under-performed when there was sparse data within some categories.

  Evaluating the performance of Bayesian cumulative logistic models in randomised controlled trials: a simulation study · 2026 · DOI

• Abstract Background Whether the magnitude of bias in randomized controlled trials (RCTs) varies systematically across study contexts remains unclear.

  Context dependence of bias in randomized controlled trials: a Bayesian meta-epidemiological study from the BFREE project · 2026 · DOI

• However, in some studies, there is limited information on the potential moderators.

  The Implications of Data Augmentation with Proxy Moderators for Generalization · 2024 · DOI

• Alternative estimation methods exist that do not require instruments, but they fail to correct for one understudied but important source of bias which we call ‘exclusion bias’.

  Exclusion Bias and the Estimation of Peer Effects · 2024 · DOI

• Future research is needed to develop, collaboratively with behavioural scientists, a suite of more robust health economic models of health-related behaviours, reported transparently, including coding, which would allow model reuse and adaptation.

  The PHEM-B toolbox of methods for incorporating the influences on Behaviour into Public Health Economic Models · 2024 · DOI

• The presence of effect modification can influence interpretation and use of study findings. The need to clarify the weighting of different strata in the total population measures. The importance of considering the impact of effect modification on policy decisions.

  Effect Modification and Its Impact on Preventable and Attributable Fractions in the Potential Outcomes Framework · 2026 · DOI

• The need for a formal justification for reporting the preventable and attributable fractions by the relevant population strata. The lack of clarity on the weighting of different strata in the total population measures.

  Effect Modification and Its Impact on Preventable and Attributable Fractions in the Potential Outcomes Framework · 2026 · DOI

• The method operates within the broader context of continuous treatment effect estimation, which presents its own inherent challenges. The need for a strong overlap condition may be violated in real-world applications with near-deterministic treatment assignments. The method can be sensitive to the size of the available dataset.

  Learning instance-specific counterfactual models for continuous treatments using hypernetworks · 2026 · DOI

Most-cited papers in Advanced Causal Inference Techniques

• An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies · Multivariate Behavioral Research · 2011 · 10,791 citations
• How Much Should We Trust Differences-In-Differences Estimates? · The Quarterly Journal of Economics · 2004 · 8,649 citations
• Asymptotic Confidence Intervals for Indirect Effects in Structural Equation Models · Sociological Methodology · 1982 · 8,207 citations
• Beyond Baron and Kenny: Statistical Mediation Analysis in the New Millennium · Communication Monographs · 2009 · 8,050 citations
• Estimating causal effects of treatments in randomized and nonrandomized studies. · Journal of Educational Psychology · 1974 · 6,578 citations
• Confidence Limits for the Indirect Effect: Distribution of the Product and Resampling Methods · Multivariate Behavioral Research · 2004 · 6,385 citations
• Identification of Endogenous Social Effects: The Reflection Problem · The Review of Economic Studies · 1993 · 4,623 citations
• Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects · American Economic Review · 2020 · 4,469 citations
• Matching Methods for Causal Inference: A Review and a Look Forward · Statistical Science · 2010 · 4,423 citations
• Estimating dynamic treatment effects in event studies with heterogeneous treatment effects · Journal of Econometrics · 2020 · 4,333 citations

Most recent work

• Starting right: aligning eligibility and treatment assignment at time zero when emulating a target trial · BMJ · 2026
• Difference-in-Differences Designs: A Practitioner’s Guide · Journal of Economic Literature · 2026
• Negative Control Falsification Tests for Instrumental Variable Designs · American Economic Review · 2026
• Medicaid Expansion and Stage at Diagnosis, Timely Initiation and Receipt of Guideline-Concordant Treatment, and Survival Among People With Non–Small Cell Lung Cancer · Journal of Clinical Oncology · 2026
• Long Story Short: Omitted Variable Bias in Causal Machine Learning · The Review of Economics and Statistics · 2026
• Causal Inferences from Digital Behavioral Data · KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie · 2026
• From prediction to intervention: causal digital twins for personalized clinical decision support · Journal of Translational Medicine · 2026
• From Bias Detection to Distributional Calibration: Negative Controls for Shared Systematic Error in Real-world Evidence Pipelines · medRxiv · 2026
• What are we estimating? Revisiting standard nutritional models through the Target Trial Framework · American Journal of Epidemiology · 2026
• The Effect of Omitted Variables on the Sign of Regression Coefficients · American Economic Review · 2026

Related topics in Mathematics