Ai Causal Inference

Foundations Of Causal Reasoning

Doctoral research examines what it means for one thing to cause another and how such claims are justified. Foundational clarity determines what any statistical procedure can legitimately establish.

Potential Outcomes Framework

Research investigates causal effects defined as comparisons between outcomes under differing treatments. This framework makes the missing data nature of causal inference explicit.

Structural Causal Models

Doctoral study addresses systems of equations representing how variables generate one another. Structural models unify prediction, intervention and counterfactual reasoning in one formalism.

Graphical Causal Models

Research examines representation of causal assumptions as diagrams of variables and arrows. Graphical representation makes assumptions visible and testable rather than implicit.

Directed Acyclic Graph Methods

Doctoral work studies acyclic diagrams encoding causal structure among measured and unmeasured variables. These diagrams determine which adjustments yield valid effect estimates.

Structural Equation Modelling

Research investigates systems of equations relating observed and latent variables. Structural equation traditions predate and inform modern causal formalism.

Interventional Distributions

Doctoral study addresses distributions arising when variables are set rather than observed. Interventional quantities differ fundamentally from conditional ones.

Counterfactual Reasoning Theory

Research examines claims about what would have happened under conditions that did not occur. Counterfactual claims require assumptions beyond those needed for interventions.

Levels Of Causal Reasoning

Doctoral work studies the distinction between association, intervention and counterfactual questions. Each level demands progressively stronger assumptions and richer models.

Identification Theory

Research investigates when a causal quantity can be recovered from available data and assumptions. Identification precedes estimation and cannot be fixed by more data.

Adjustment Criteria In Graphs

Doctoral study addresses graphical rules determining which variable sets permit valid adjustment. These criteria replaced informal reasoning about which variables to control.

Calculus Of Interventions

Research examines formal rules transforming interventional expressions into observational ones. This calculus is complete for identification from observational data.

Transportability Of Causal Findings

Doctoral work studies transfer of causal conclusions between populations and settings. Transport requires explicit assumptions about which mechanisms are shared.

External Validity Theory

Research investigates whether findings from a study population apply elsewhere. External validity is frequently assumed rather than formally examined.

Assumptions Of Conditional Independence

Doctoral study addresses the requirement that treatment is effectively random given adjustment. This assumption is untestable and underlies most observational analysis.

Positivity And Overlap Conditions

Research examines the requirement that every unit could plausibly receive each treatment. Violations of this condition produce estimates driven by extrapolation.

Consistency And Well Defined Interventions

Doctoral work studies whether the treatment being estimated is sufficiently precisely specified. Vaguely defined exposures yield effects with no clear interpretation.

Interference Between Units

Research investigates situations where one unit treatment affects another unit outcome. Interference invalidates the independence assumption most methods require.

Spillover Effect Modelling

Doctoral study addresses estimation of effects transmitted between connected individuals. Spillovers are frequently the effect of greatest policy interest.

Estimands And Their Definition

Research examines precise specification of the quantity an analysis aims to estimate. Ambiguous estimands cause disagreement that no statistical method resolves.

Average Treatment Effect Estimation

Doctoral work studies estimation of the mean effect across a defined population. This quantity remains the default target of most causal analyses.

Conditional Treatment Effect Estimation

Research investigates effects varying with observed characteristics of individuals. Conditional effects support targeting rather than uniform policy.

Individual Treatment Effect Estimation

Doctoral study addresses effects for specific individuals rather than groups. Individual effects are never observed and require strong additional assumptions.

Quantile Treatment Effects

Research examines effects at different points of the outcome distribution. Mean effects can conceal harm concentrated in part of a population.

Distributional Causal Effects

Doctoral work studies how interventions change entire outcome distributions. Distributional targets capture inequality effects that means obscure entirely.

Mediation Analysis Theory

Research investigates decomposition of effects into pathways through intermediate variables. Mediation questions require assumptions considerably stronger than total effects.

Path Specific Effects

Doctoral study addresses effects transmitted along particular routes through a causal structure. Path specific quantities are central to discrimination and mechanism analysis.

Moderation And Effect Modification

Research examines how effects differ across subgroups defined by other variables. Effect modification is frequently confused with statistical interaction.

Causal Attribution Methods

Doctoral work studies attributing observed outcomes to particular prior causes. Attribution questions are backward looking and distinct from effect estimation.

Probability Of Causation

Research investigates the chance that a specific outcome was caused by a specific exposure. This quantity is central to legal and compensation contexts.

Propensity Score Methods

Doctoral study addresses summarising many covariates into a single treatment probability. Propensity summarisation reduces high dimensional adjustment to one dimension.

Matching Estimators

Research examines pairing treated and untreated units with similar characteristics. Matching makes comparisons transparent but discards unmatched observations.

Inverse Probability Weighting

Doctoral work studies reweighting observations to construct a pseudo randomised population. Weighting handles time varying treatment that adjustment cannot.

Stabilised Weighting Approaches

Research investigates weight construction reducing variance while preserving validity. Unstabilised weights produce estimates dominated by a few observations.

Doubly Robust Estimation

Doctoral study addresses estimators consistent if either outcome or treatment model is correct. Double robustness provides protection against a single modelling failure.

Targeted Learning Approaches

Research examines estimation combining flexible learning with formal statistical guarantees. Targeted approaches optimise directly for the causal quantity of interest.

Augmented Weighting Estimators

Doctoral work studies combination of weighting and outcome modelling in one estimator. Combination improves efficiency beyond either component alone.

Influence Function Based Estimation

Research investigates estimators derived from the mathematical sensitivity of a quantity. Influence functions provide the foundation for modern efficient estimation.

Semiparametric Efficiency Theory

Doctoral study addresses the best achievable precision under minimal modelling assumptions. Efficiency bounds indicate whether an estimator can be improved further.

Regression Adjustment Methods

Research examines effect estimation through modelling outcomes given treatment and covariates. Regression adjustment remains ubiquitous and frequently misapplied.

Outcome Modelling Strategies

Doctoral work studies specification of models relating covariates to outcomes. Model misspecification propagates directly into effect estimates.

Machine Learning For Nuisance Estimation

Research investigates flexible models for components not of direct interest. Flexible nuisance estimation introduces bias unless handled carefully.

Cross Fitting And Sample Splitting

Doctoral study addresses separating data used for nuisance estimation and effect estimation. Splitting removes bias arising from overfitting nuisance components.

Debiased Machine Learning

Research examines frameworks combining flexible learning with valid statistical inference. These frameworks made machine learning usable for effect estimation.

Causal Forest Methods

Doctoral work studies tree ensembles adapted to estimate varying treatment effects. Forest methods discover effect variation without prespecifying subgroups.

Tree Based Heterogeneity Estimation

Research investigates recursive partitioning to identify subgroups with differing effects. Tree methods produce interpretable subgroup definitions directly.

Meta Learner Approaches

Doctoral study addresses strategies decomposing effect estimation into prediction problems. Meta learners permit any prediction method to be reused for causal work.

Representation Learning For Causal Effects

Research examines learned representations balancing treated and untreated groups. Learned balancing handles high dimensional covariates such as text and images.

Neural Network Based Effect Estimation

Doctoral work studies deep architectures designed for treatment effect estimation. Neural approaches suit settings with very complex covariate structure.

Generative Models In Causal Estimation

Research investigates generative approaches to modelling counterfactual outcomes. Generative models can represent uncertainty about unobserved outcomes explicitly.

Bayesian Causal Inference

Doctoral study addresses causal estimation within a probabilistic inferential framework. Bayesian treatment expresses uncertainty about both effects and assumptions.

Prior Specification In Causal Models

Research examines how prior beliefs enter causal estimation and affect conclusions. Prior choice matters most precisely where data is least informative.

Nonparametric Bayesian Causal Methods

Doctoral work studies flexible probabilistic models with unbounded complexity for causal work. These models avoid committing to a fixed functional form.

Uncertainty Quantification For Effects

Research investigates honest expression of confidence in estimated causal quantities. Reported precision frequently ignores uncertainty in the assumptions themselves.

Confidence Interval Construction

Doctoral study addresses interval estimation for causal quantities under flexible modelling. Valid intervals are considerably harder to construct than point estimates.

Multiple Testing In Causal Analysis

Research examines error control when many effects are estimated simultaneously. Subgroup and multi outcome analyses inflate false discovery substantially.

Finite Sample Behaviour Of Estimators

Doctoral work studies estimator performance at realistic rather than asymptotic sample sizes. Asymptotic guarantees frequently fail at the sample sizes actually available.

High Dimensional Confounding Adjustment

Research investigates adjustment when covariates vastly outnumber observations. High dimensional settings defeat conventional adjustment approaches entirely.

Variable Selection For Adjustment

Doctoral study addresses which variables should and should not be adjusted for. Adjusting for the wrong variables introduces rather than removes bias.

Regularisation In Causal Estimation

Research examines penalised estimation applied to causal rather than predictive targets. Regularisation biases estimates in ways that require correction.

Overlap Diagnostics And Trimming

Doctoral work studies detection and handling of regions with insufficient comparison data. Trimming changes the population to which estimates actually apply.

Handling Of Extreme Weights

Research investigates observations receiving disproportionate influence in weighted estimation. Extreme weights produce unstable estimates with misleading precision.

Balance Assessment Methods

Doctoral study addresses checking whether adjustment achieved comparable groups. Balance diagnostics are the primary check available in observational analysis.

Covariate Balancing Approaches

Research examines methods estimating weights to achieve balance directly. Direct balancing avoids the model misspecification propensity estimation risks.

Entropy Balancing Methods

Doctoral work studies weighting schemes satisfying balance constraints with minimal weight variation. These methods guarantee balance on specified moments exactly.

Synthetic Control Methods

Research investigates constructing a comparison unit from a weighted combination of others. Synthetic controls suit settings with a single treated unit.

Augmented Synthetic Control

Doctoral study addresses extensions correcting bias in synthetic control estimation. Augmentation addresses imperfect fit during the pretreatment period.

Difference In Differences Estimation

Research examines comparison of changes over time between treated and untreated groups. This design is among the most widely used in policy evaluation.

Testing Of Parallel Trends

Doctoral work studies the assumption that groups would have moved together absent treatment. This assumption is untestable yet routinely assessed informally.

Staggered Adoption Designs

Research investigates settings where units are treated at different times. Staggered timing invalidates the standard estimators applied for decades.

Critiques Of Fixed Effects Estimation

Doctoral study addresses recently identified problems with conventional panel estimators. These critiques prompted substantial methodological revision across social science.

Event Study Designs

Research examines effects estimated relative to the timing of an intervention. Event studies reveal dynamics that single effect estimates conceal.

Regression Discontinuity Designs

Doctoral work studies effects identified at a threshold determining treatment assignment. Discontinuity designs approach randomisation near the threshold.

Bandwidth Selection In Discontinuity Designs

Research investigates how much data around the threshold should be used. Bandwidth choice trades bias against precision and drives conclusions.

Fuzzy Discontinuity Approaches

Doctoral study addresses thresholds affecting treatment probability rather than determining it. Fuzzy designs combine discontinuity and instrumental reasoning.

Instrumental Variable Methods

Research examines identification using variables affecting treatment but not outcomes directly. Instruments address confounding that no adjustment could remove.

Weak Instrument Problems

Doctoral work studies behaviour when instruments explain little treatment variation. Weak instruments produce badly biased estimates with misleadingly narrow intervals.

Instrument Validity Testing

Research investigates assessment of whether instrument assumptions plausibly hold. Instrument validity is fundamentally untestable and must be argued substantively.

Mendelian Randomisation

Doctoral study addresses use of genetic variants as instruments for exposures. Genetic instruments are fixed at conception and precede most confounding.

Genetic Instrument Selection

Research examines choice of variants and handling of their multiple effects. Variants affecting several traits violate the core instrumental assumption.

Randomised Experiment Design

Doctoral work studies design of experiments providing the strongest causal evidence. Design decisions determine what questions the experiment can answer.

Cluster Randomised Designs

Research investigates randomisation of groups rather than individuals. Clustering reduces statistical power and introduces distinctive analytical requirements.

Stepped Wedge Designs

Doctoral study addresses sequential rollout of an intervention across units over time. This design suits interventions that must eventually reach everyone.

Factorial Experimental Designs

Research examines experiments varying several factors simultaneously. Factorial designs estimate interactions that separate experiments cannot.

Sequential Multiple Assignment Trials

Doctoral work studies trials randomising at several decision points in sequence. These designs estimate treatment strategies rather than single treatments.

Adaptive Experimental Design

Research investigates experiments modifying allocation as results accumulate. Adaptation improves efficiency and participant experience simultaneously.

Online Controlled Experiments

Doctoral study addresses continuous experimentation in deployed digital systems. These experiments run at scales conventional trials cannot approach.

Experiment Platform Design

Research examines infrastructure supporting large volumes of concurrent experiments. Platform design determines both experiment validity and organisational throughput.

Noncompliance And Treatment Receipt

Doctoral work studies participants not receiving the treatment assigned to them. Noncompliance separates the effect of assignment from the effect of treatment.

Principal Stratification

Research investigates effects within groups defined by their behaviour under both treatments. Principal strata are latent and require careful assumptions to address.

Intercurrent Event Handling

Doctoral study addresses events occurring after treatment that complicate outcome interpretation. Handling strategy defines the estimand rather than merely the analysis.

Attrition And Missing Outcomes

Research examines participants who leave a study and whose outcomes are never observed. Loss of participants is rarely unrelated to the outcome the study aims to measure.

Missing Data Under Causal Assumptions

Doctoral work studies incomplete data treated within a causal framework. Missingness mechanisms can be represented explicitly as part of the causal structure.

Measurement Error In Causal Analysis

Research investigates consequences of imprecisely measured exposures and outcomes. Measurement error can either weaken or exaggerate estimated effects.

Misclassification Of Exposure

Doctoral study addresses incorrect categorisation of who received a treatment. Differential misclassification can create effects where none exist.

Selection Bias Mechanisms

Research examines distortion arising from how units enter the analysed sample. Selection can be represented and sometimes corrected within causal graphs.

Collider Bias Analysis

Doctoral work studies bias created by conditioning on common consequences. Collider bias is counterintuitive and appears throughout applied research.

Immortal Time Bias

Research investigates bias from periods during which outcomes could not occur. This bias has produced spectacular false findings in observational medicine.

Confounding By Indication

Doctoral study addresses treatment assignment driven by prognosis itself. This confounding is the central difficulty in observational treatment comparison.

Time Varying Confounding

Research examines confounding variables that are themselves affected by earlier treatment. Conventional regression adjustment fails entirely in this common longitudinal situation.

Marginal Structural Models

Doctoral work studies models estimating effects of treatment sequences using weighting. These models handle time varying confounding correctly.

Sequential Regression Estimation

Research investigates iterated outcome modelling to estimate effects of treatment strategies. This approach complements weighting for longitudinal problems.

Structural Nested Models

Doctoral study addresses models specifying effects of treatment at each time given history. These models estimate effects of continuous and repeated treatment.

Longitudinal Causal Inference

Research examines causal analysis with repeated measurement over extended periods. Longitudinal structure creates both opportunity and severe complication.

Dynamic Treatment Regimes

Doctoral work studies rules assigning treatment based on evolving patient state. Regimes represent the sequential decisions clinical practice actually involves.

Optimal Regime Estimation

Research investigates learning treatment rules maximising expected outcomes. Optimal regimes personalise treatment based on individual characteristics and history.

Survival Analysis Under Causal Framing

Doctoral study addresses time to event outcomes within causal frameworks. Censoring interacts with confounding in ways requiring careful treatment.

Competing Risks In Causal Analysis

Research examines settings where several event types preclude one another. Competing events complicate the definition of the causal quantity itself.

Outcomes Unavailable After Death

Doctoral work studies outcomes undefined for units that did not survive. This situation requires estimands that conventional analysis cannot express.

Recurrent Event Causal Analysis

Research investigates effects on events that occur repeatedly within individuals. Recurrent events carry information that time to first event discards.

Panel Data Causal Methods

Doctoral study addresses causal analysis with repeated observation of the same units. Panel structure permits control for stable unobserved characteristics.

Time Series Causal Analysis

Research examines causal questions in single long sequences of observations. Temporal ordering provides information that cross sectional data lacks.

Temporal Predictive Causal Testing

Doctoral work studies whether past values of one series help predict another. Predictive testing is widely used yet frequently misinterpreted as causal proof.

Intervention Analysis In Time Series

Research investigates the effects of discrete events on the behaviour of ongoing time series. This approach evaluates policies and shocks introduced at precisely known moments.

Spatial Causal Inference

Doctoral study addresses causal analysis where units are located in space. Spatial proximity creates both confounding and interference simultaneously.

Causal Structure Discovery

Research examines inference of causal relationships from data rather than assumption. Discovery attempts what most methods treat as given beforehand.

Constraint Based Discovery Algorithms

Doctoral work studies structure learning from patterns of conditional independence. These algorithms recover structure up to an equivalence class only.

Score Based Structure Learning

Research investigates search over structures optimising a goodness measure. Score based search scales differently than constraint based approaches.

Continuous Optimisation For Structure Learning

Doctoral study addresses reformulation of structure search as continuous optimisation. This reformulation permits gradient methods to be applied directly.

Functional Causal Model Discovery

Research examines direction identification using assumptions about functional form. Functional assumptions break symmetries that independence testing cannot.

Additive Noise Model Methods

Doctoral work studies direction discovery assuming noise enters additively. These methods identify direction between two variables without intervention.

Discovery With Latent Confounders

Research investigates structure learning when important variables are unmeasured. Unmeasured confounding is the normal rather than exceptional condition.

Discovery From Interventional Data

Doctoral study addresses structure learning using data from performed interventions. Interventional data resolves ambiguities observational data cannot.

Active Intervention Selection

Research examines choosing which interventions to perform for maximum structural information. Experiments are costly and their selection should be principled.

Discovery In Time Series Data

Doctoral work studies structure learning exploiting temporal ordering of observations. Time ordering constrains direction without further assumptions.

Discovery Evaluation And Benchmarking

Research investigates fair assessment and comparison of structure learning algorithms. Evaluation is inherently difficult because the true causal structure is rarely known.

Scalability Of Discovery Algorithms

Doctoral study addresses causal structure learning across very large numbers of variables. The space of possible structures grows faster than exponentially with variable count.

Prior Knowledge In Structure Learning

Research examines incorporation of domain knowledge into discovery algorithms. Partial knowledge substantially constrains the space that must be searched.

Causal Feature Selection

Doctoral work studies identification of variables with genuine causal relevance. Causal features transfer across settings where predictive features do not.

Sensitivity Analysis For Confounding

Research investigates how strong unmeasured confounding would need to be to overturn conclusions. Sensitivity analysis converts an untestable assumption into a quantitative statement.

Bounding Approaches For Effects

Doctoral study addresses ranges of possible effects consistent with weaker assumptions. Bounds provide credible conclusions where point identification fails.

Partial Identification Methods

Research examines inference when assumptions identify a set rather than a value. Partial identification trades precision for credibility explicitly.

Negative Control Methods

Doctoral work studies auxiliary variables used to detect residual confounding. Negative controls provide an empirical check on untestable assumptions.

Proximal Causal Inference

Research investigates identification using proxies for unmeasured confounders. Proximal methods recover effects where conventional adjustment cannot.

Falsification Testing Of Assumptions

Doctoral study addresses empirical tests capable of refuting stated causal assumptions. Causal assumptions can never be verified, but a useful subset can be falsified by data.

Placebo And Refutation Tests

Research examines analyses that should show no effect if assumptions hold. Refutation tests are among the most practical assumption checks available.

Triangulation Across Designs

Doctoral work studies combining designs with different assumptions and biases. Agreement across designs with unrelated weaknesses strengthens conclusions substantially.

Benchmarking Against Randomised Evidence

Research investigates whether observational methods reproduce randomised trial results. Benchmarking is the strongest available validation of observational methods.

Simulation Studies In Causal Methods

Doctoral study addresses evaluation of methods on data with known causal structure. Simulation is the primary route to understanding when methods fail.

Synthetic Benchmark Construction

Research examines construction of realistic test problems with known answers. Benchmark realism determines whether evaluation predicts real performance.

Reproducibility In Causal Analysis

Doctoral work studies whether published causal findings can be independently regenerated. Analytical flexibility in causal work is unusually large.

Software Development For Causal Methods

Research investigates implementation of causal methods in usable software. Software availability determines which methods practitioners actually apply.

Preregistration Of Observational Analyses

Doctoral study addresses advance specification of observational analysis plans. Preregistration limits the analytical flexibility that produces false findings.

Reporting Standards For Causal Claims

Research examines what should be disclosed when causal conclusions are published. Reporting standards make assumptions visible to readers and reviewers.

Meta Analysis Of Causal Estimates

Doctoral work studies quantitative synthesis of effect estimates across studies. Synthesis must account for differing estimands and assumptions.

Evidence Synthesis Across Designs

Research investigates combining evidence from experimental and observational sources. Combination requires reconciling different assumptions and target populations.

Publication Bias In Causal Literature

Doctoral study addresses selective publication and reporting of causal research findings. Selective publication systematically distorts the apparent strength of causal evidence.

Replication Of Causal Findings

Research examines whether causal conclusions hold in independent data. Replication failures are common and frequently attributed to context.

Robustness Checking Practice

Doctoral work studies how analysts assess stability of conclusions to analytical choices. Selective reporting of robustness checks is itself a source of bias.

Specification Curve Analysis

Research investigates systematic reporting across many defensible analytical specifications. This approach exposes how much conclusions depend on analyst choices.

Causal Representation Learning

Doctoral study addresses learning representations corresponding to causal variables. Causal representations promise generalisation that predictive ones lack.

Disentanglement And Causal Factors

Research examines separating independent generative factors within learned representations. Disentanglement is closely connected to causal identifiability.

Invariant Prediction Methods

Doctoral work studies prediction using relationships stable across environments. Invariance provides a route to causal structure without intervention.

Out Of Distribution Generalisation

Research investigates model performance when deployment conditions differ from training. Causal structure explains which relationships should be expected to persist.

Domain Adaptation Through Causal Structure

Doctoral study addresses transferring models between settings using causal assumptions. Causal reasoning specifies what must be adjusted when domains differ.

Robustness To Distribution Shift

Research examines the stability of model performance when input distributions change. Causal relationships remain valid under shifts that break purely correlational ones.

Causal Approaches To Fairness

Doctoral work studies fairness definitions expressed in causal rather than statistical terms. Causal framing distinguishes discrimination from mere association.

Counterfactual Fairness Definitions

Research investigates fairness requiring outcomes unchanged in a counterfactual world. These definitions make explicit the mechanisms considered unacceptable.

Discrimination Analysis Through Causal Models

Doctoral study addresses legal and social discrimination questions using causal formalism. Causal decomposition separates permitted from prohibited pathways.

Explainability Through Counterfactuals

Research examines explanation of model behaviour by describing minimal changes. Counterfactual explanations are intuitive and actionable for affected individuals.

Counterfactual Explanation Generation

Doctoral work studies algorithms producing plausible and useful counterfactual explanations. Generated explanations must be feasible for the individual to act upon.

Recourse And Actionable Explanation

Research investigates whether individuals can realistically achieve a different outcome. Recourse connects explanation to genuine agency rather than description.

Causal Reinforcement Learning

Doctoral study addresses sequential decision learning informed by causal structure. Causal knowledge improves both efficiency and transfer in decision learning.

Off Policy Evaluation Methods

Research examines estimating the value of a policy from data generated by another. Off policy evaluation is a causal problem in decision making form.

Policy Learning From Observational Data

Doctoral work studies deriving decision rules from data that was not experimental. Policy learning inherits every assumption effect estimation requires.

Contextual Bandit Causal Methods

Research investigates sequential decisions with immediate feedback and confounding. Bandit settings combine estimation with active data collection.

Exploration Under Confounding

Doctoral study addresses learning when past decisions were driven by unobserved information. Confounded logs mislead learners that assume random exploration.

Causal Approaches In Recommendation

Research examines recommendation framed as intervention rather than prediction. Recommendation systems influence the very behaviour they aim to predict.

Uplift Modelling

Doctoral work studies prediction of the incremental effect of an action per individual. Uplift targets who to act upon rather than what will happen.

Targeting And Assignment Policy

Research investigates allocation of limited treatment to those who benefit most. Optimal targeting requires effect heterogeneity to be reliably estimated.

Causal Effects Of Language Model Outputs

Doctoral study addresses measuring how generated content changes downstream behaviour. Deployed language systems intervene rather than merely describe.

Causal Evaluation Of Model Behaviour

Research examines intervening on models to establish what drives their outputs. Causal evaluation distinguishes genuine mechanisms from correlated artefacts.

Mechanistic Interpretability Through Intervention

Doctoral work studies internal model mechanisms identified by direct manipulation. Intervention establishes causal roles that correlational probing cannot.

Activation Patching Methods

Research investigates substituting internal model states to trace information flow. Patching localises which components carry which information causally.

Causal Abstraction Of Models

Doctoral study addresses whether a simple causal model faithfully describes a complex one. Abstraction provides a principled account of what a model implements.

Model Editing Through Causal Analysis

Research examines targeted modification of model behaviour using causal localisation. Editing requires knowing which components genuinely carry a behaviour.

Causal Inference For Simulation Models

Doctoral work studies causal questions posed within complex computer simulations. Simulations permit intervention that observational systems do not.

Digital Experimentation At Scale

Research investigates the methodology of very large scale online experimentation. Scale introduces problems of interference, heterogeneity and multiplicity.

Causal Inference In Interactive Systems

Doctoral study addresses causal analysis where the system and users adapt to each other. Mutual adaptation violates the stability most methods assume.

Agent Based Causal Analysis

Research examines causal reasoning within simulations of interacting agents. Agent models make interference and feedback explicit rather than assumed away.

Causal Inference In Epidemiology

Doctoral work studies causal methods applied to disease occurrence and prevention. Epidemiology originated much of the modern causal inference framework.

Comparative Effectiveness Research

Research investigates comparison of treatments as used in routine practice. Effectiveness questions concern real use rather than idealised trial conditions.

Target Trial Emulation

Doctoral study addresses designing observational analyses to mimic a hypothetical trial. Explicit emulation prevents several common and severe design errors.

Causal Methods In Health Policy

Research examines evaluation of health system interventions and reforms. Policy interventions are rarely randomised yet require causal evaluation.

Causal Inference In Economics

Doctoral work studies causal methodology as developed within economics. Economic methodology contributed instruments, discontinuities and difference designs.

Policy Evaluation Methods

Research investigates assessment of whether policies achieved their intended effects. Evaluation credibility determines whether evidence influences policy at all.

Labour Market Causal Studies

Doctoral study addresses effects of policies and shocks on employment and earnings. Labour economics provided many foundational causal design innovations.

Causal Inference In Education Research

Research examines effects of educational interventions and policies. Educational settings involve clustering, interference and long delayed outcomes.

Causal Analysis In Social Science

Doctoral work studies causal questions about social processes and institutions. Social systems involve feedback that complicates standard causal framing.

Causal Inference In Political Science

Research investigates causal effects of institutions, campaigns and policies. Political settings rarely permit experimentation on questions of interest.

Causal Methods In Environmental Science

Doctoral study addresses causal attribution in environmental and ecological systems. Environmental systems involve spatial dependence and long time scales.

Climate Attribution Methods

Research examines attribution of specific events to changing climate conditions. Attribution combines physical modelling with statistical causal reasoning.

Causal Inference In Genomics

Doctoral work studies causal effects of genetic variation on traits and disease. Genetic settings offer both natural experiments and severe confounding.

Single Cell Perturbation Analysis

Research investigates causal effects inferred from experiments perturbing individual cells. Perturbation experiments provide interventional data at very large scale.

Causal Inference In Neuroscience

Doctoral study addresses causal relationships among neural activity and behaviour. Neural systems permit intervention that most biological systems do not.

Causal Discovery In Systems Biology

Research examines inference of regulatory structure from molecular measurements. Biological networks involve feedback that acyclic methods cannot represent.

Philosophy Of Causation

Doctoral work studies philosophical accounts of what causation actually is. Philosophical analysis clarifies disputes that statistics alone cannot settle.

Causal Explanation And Understanding

Research investigates what makes a causal account explanatory rather than merely correct. Explanation involves conveying mechanism beyond stating an effect.

Teaching Of Causal Inference

Doctoral study addresses how causal reasoning is taught and where learners struggle. Widespread confusion between association and causation persists after training.

Ethics Of Causal Claims In Policy

Research examines responsibilities arising when causal claims influence public decisions. Overstated causal claims have produced substantial and documented harm.