Causal Inference for Case Studies in Behavioral Health

Shane Sparkes

REVIEW 3 major objections 5 minor 45 references

Causal Inference for Case Studies in Behavioral Health

T0 review · 3 major / 5 minor · reviewed 2026-07-09 · glm-5.2

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

Pith's one-line read Support, not distribution: causal inference without measuring confounders

desk verdict Support-based causal estimands identified under positivity alone — correct but practically limited read the letter →

arxiv 2607.06912 v1 pith:SC5UBVGG submitted 2026-07-08 stat.ME

Shane Sparkes This is my paper

classification stat.ME

keywords causalframeworkbehavioralcasecontrastsestimandsbaselineconfounding

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 introduces a class of causal estimands called Omega estimands, which are defined as contrasts of functions of an outcome variable's support (the set of values it can take) rather than its probability distribution. The central claim is that, in a structural causal model with a treatment T, outcome Y, and unmeasured confounders U, the observational support of Y given T equals the interventional support of Y under do(T) provided only that positivity holds, meaning every treatment value remains possible under every confounder value. Because support sets do not encode how probability mass is distributed across them, confounders that would normally bias distribution-based estimands (like mean differences) leave support-based estimands untouched. The paper proves this coincidence (Proposition 1), then extends the framework with two optional conditions: one allowing a client's recalled baseline to substitute for sparsely measured pre-treatment data, and one linking support contrasts back to conventional mean contrasts via an identity between the functional average of a support and the expected value. The framework is situated in behavioral health case studies, where providers cannot measure all confounders and cannot randomize or withhold treatment, but still need to reason about whether a given course of care changed what outcomes were possible for a client.

What carries the argument

The key objects are Omega estimands: contrasts of functions (min, max, range, cardinality, functional average) applied to the support of an outcome variable, rather than its distribution. The central result (Proposition 1) shows that in a DAG with nodes {Y, T, U} and edges T→Y, U→T, U→Y, the observational support Omega_{Y|t} equals the interventional support Omega_{Y|do(t)} under positivity f(t|u) > 0 for all t, u. The proof compares f(y|t) = sum_u f(y|t,u)f(u|t) with f(y|do(t)) = sum_u f(y|t,u)f(u) and shows both are positive on exactly the same set of y values because positivity forces f(u) and f(u|t) to share the same support over Omega_U. Two auxiliary conditions extend the framework: ep

What would settle it

If, in a substantial fraction of real behavioral health treatment episodes, the support of outcome variables does not change under effective treatment, then Omega estimands would be formally identified but substantively uninformative, rendering the framework correct but empty for most practical purposes.

Watch

Extended reading notes

Core claim

Under a standard causal graph with treatment T, outcome Y, and unmeasured confounders U, the support of Y observed under T equals the support Y would have under experimental intervention do(T), as long as positivity holds, that is, every treatment value remains possible for every value of the confounders. This means contrasts defined on supports, rather than on probability distributions, are causally identified without measuring, knowing, or adjusting for any confounders. The proof hinges on the fact that the observational density f(y|t) and the interventional density f(y|do(t)) differ only in how they weight confounder strata, via f(u|t) versus f(u), but positivity ensures both weights are非

Load-bearing premise

The framework is only substantively useful if outcome supports, understood as the set of values a client considers possible, actually shift under treatment. If a treatment changes only how likely each value is, without eliminating or adding any possible values, Omega estimands detect nothing, even though the treatment may be genuinely effective.

Editorial extensions

If this is right

Providers in behavioral health settings could assess treatment efficacy for individual clients without needing to identify or measure confounders, using only the set of outcome values considered possible before and after treatment.
Program evaluation in population-based studies could target Omega effects (support contrasts) as complementarity or alternative estimands, gaining identification under positivity alone once a population frame is granted.
The expected-average identity Av(Y) = E(Y) + kappa provides a bridge where conventional mean-difference estimands acquire causal meaning under milder conditions than standard backdoor adjustment, if the kappa parameter is preserved or bounded across treatment conditions.
Sensitivity analysis for unmeasured confounding could be reframed: instead of asking how strong confounding must be to explain away a mean difference, one asks whether confounding could plausibly change which outcome values are possible, which is a different and potentially more tractable question.

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.

Desk Editor's Note

read the letter

Here's the short version: this paper proves that under positivity alone, the observational and interventional *supports* of an outcome variable coincide — no backdoor adjustment needed. The proof is clean, correct, and self-contained. The trade-off is that you're identifying support-based contrasts (min, max, range, etc.), not distributional ones, so the framework only detects treatment effects that change which outcome values are possible, not effects that merely shift probability mass around. That's a real limitation and the author is honest about it in Section 5. The stress-test note flags that positivity is stronger than the paper lets on — it requires f(t|u)>0 for ALL values of unmeasured confounders U, which is unverifiable. The paper's own crisis example shows how easily it fails. The defense is that coarse treatment conceptualizations (binary exposure over a period) salvage it, but the coarser T gets, the less clinically informative the contrast. This tension is real and unresolved. The de Finetti framing doesn't fix it: even under subjective probability, the agent must believe treatment is possible under every confounder value, which is a substantive constraint, not just a coherence requirement. That said, the stress-test slightly overstates the problem. The paper does acknowledge positivity is untestable and must be argued case by case via the DAG. It's not hiding the issue — it just doesn't fully resolve the granularity trade-off. The extensions (Conditions II and III) are weaker. Condition II links support contrasts to mean contrasts via the expected-average identity Av(Y) = E(Y) + κ, which rests on the author's own prior work (Sparkes et al., 2024) without independent validation. Condition III (support recall) is milder than the recall assumptions already baked into instruments like the PHQ-9, so it's defensible but still untestable. The data application is purely illustrative — subjective priors and likelihoods with no empirical grounding. It shows the mechanics but proves nothing about real-world utility. The core result (Proposition 1) is genuinely novel for the SCM literature, which focuses on distributional identification. Whether support-based estimands are substantively useful in behavioral health practice is an open empirical question the paper flags but doesn't answer. This deserves a serious referee. The math is sound, the framing is original, and the limitations are mostly acknowledged. A referee should push hard on the positivity–granularity trade-off and demand more on when supports actually shift in practice.

Referee Report

3 major / 5 minor

Summary. The manuscript proposes a framework for causal inference in observational N=1 settings (case studies), motivated by behavioral health care. The central idea is to define causal estimands—termed Ω estimands—as contrasts of functions of an outcome variable's support rather than its distribution. The main theoretical result (Proposition 1) proves that under a standard DAG {Y, T, U} with edges T→Y, U→T, U→Y, and a positivity assumption, the observational and interventional supports of the outcome coincide (Ω_{Y|t} = Ω_{Y|do(t)}). This renders Ω estimands identified without measuring or adjusting for confounders. Two optional conditions extend the framework: 'support recall' (Condition III), which licenses a client's recalled baseline as a stand-in for unmeasured baseline periods, and 'epistemic preservation' (Condition II), which connects support contrasts to mean contrasts via an expected-average identity (Av(Y) = E(Y) + κ). A subjectivist (de Finetti) interpretation of probability is adopted throughout. A case study of CBT for anxiety illustrates a simple Bayesian application.

Significance. The paper addresses a genuine methodological gap: causal inference for individual care processes under pervasive unmeasured confounding, where standard adjustment and matching strategies are infeasible. Proposition 1 is a clean, self-contained result: the proof correctly shows that because positivity ensures f(u) and f(u|t) are both strictly positive on all of Ω_U, the mixture representations f(y|do(t)) = Σ_u f(y|t,u)f(u) and f(y|t) = Σ_u f(y|t,u)f(u|t) are positive on exactly the same set of y-values, establishing support equality. The idea that support-based estimands sidestep confounding by discarding distributional information is conceptually interesting and potentially useful for program evaluation beyond N=1 settings, as the author notes. The de Finetti framing is philosophically coherent and gives the support concept a subjective interpretation that is natural for clinical reasoning. The expected-average identity and the U random variable class, while drawn from the author's prior work, provide a concrete bridge between support contrasts and conventional mean contrasts.

major comments (3)

Section 3 (Proposition 1) and Section 5 (Discussion): The central tension between positivity and clinical meaningfulness is acknowledged but not resolved. The paper correctly notes that positivity is easiest to satisfy when treatment is 'coarsely conceptualized' (e.g., binary T=1 if any case management occurred). However, the coarser T becomes, the less informative the causal contrast is for clinical decision-making. The paper's own counterexample—a client in crisis precluding casual interventions—illustrates how easily positivity fails at granular levels. The author asserts that 'certain coarse conceptualizations of clinical intervention do fulfill positivity' but provides no general criterion for when this holds. This is load-bearing because the entire framework's applicability rests on positivity being defensible. The author should either (a) provide a more systematic characterization
Section 5 (Discussion), paragraph on the 'obvious objection': The paper acknowledges that 'treatments that alter only the distribution of probability over an unchanged set of possibilities are invisible to Ω effects.' This is a substantive limitation that threatens the framework's practical utility. If supports rarely change in practice, the framework is formally correct but substantively empty. The de Finetti interpretation partially addresses this by making supports subjective (a client's lived possibilities can shift even if the data-generating support does not), but this move is philosophical rather than empirical. The paper would benefit from either empirical evidence or a more rigorous argument that support shifts are common enough in behavioral health settings to make the framework practically useful. Without this, the framework's applicability remains an open question rather than
Section 4 (Data Application): The Bayesian calculation Pr(θ|y0,y*) = 0.5895 relies on entirely subjective specifications: Pr(θ) = 0.35, Pr(y0,y*|θ) = 0.8, Pr(y0,y*|θc) = 0.3. While the author states these are illustrative, the example is presented as demonstrating the framework's clinical applicability. The sensitivity analysis (varying the prior from 0.20 to 0.50) shows posteriors between 0.40 and 0.73, but the likelihoods are held fixed at values that strongly favor θ. The example would be more convincing if it demonstrated how a provider would arrive at these likelihood values from clinical reasoning, or if it showed sensitivity to the likelihood specifications as well. As presented, the example illustrates the mechanics of Bayes' theorem rather than the framework's ability to generate credible causal inferences from clinical data.

minor comments (5)

Section 2.1: The distinction between a random variable's 'scale' and its 'support' is important but introduced somewhat obliquely. A clearer definition upfront would help readers avoid the 'faulty equivocation of scales and supports' the author warns about.
Section 3, Condition II: The term 'epistemic preservation' is somewhat opaque. Consider a more descriptive name, or at least a brief parenthetical explaining the condition's content when first introduced.
Section 3, Proposition 2: The formula for κ involves f(M-i) notation that could be clearer. Explicitly stating that f(M-i) denotes the probability of the i-th value below the maximum would aid readability.
Section 4: The data sequence {10,9,9,...,4} is described as 'anxiety scores' but the scale (0-10) is only mentioned in passing. Clarifying the measurement instrument and scoring upfront would help.
The reference to 'Sparkes et al. (2024)' for the expected-average identity and U random variables is frequent. Given that these are load-bearing for Condition II and the extension to mean contrasts, the paper would benefit from a self-contained derivation or at least a more detailed summary of the relevant results, rather than frequent deferral to the cited work.

Circularity Check

1 steps flagged · score 2.0 of 10

Main proposition is self-contained; self-citation supports only an optional extension and is itself a mathematical identity

self citation load bearing [Section 3, Condition II and the expected-average identity]
"A useful identity is as follows (Sparkes et al., 2024): Av(Y) = E(Y) + κ. We call it the expected-average identity. Thus κ = Av(Y) − E(Y), and |κ| measures the distance between the functional average and the expected value."

The expected-average identity Av(Y) = E(Y) + κ is cited to Sparkes et al. (2024), where the present author is a co-author. However, examining the definition of κ = R^{-1} σ_{X,f(X)^{-1}} and working through the discrete case (R = |Ω|): κ = |Ω|^{-1} Σ_y y/f(y)·f(y) − E(Y) = |Ω|^{-1} Σ_y y − E(Y) = Av(Y) − E(Y). So the identity is a mathematical tautology given the definition of κ — it holds by construction. The self-citation is to the paper where this was first written down, not to an unverified empirical claim. This is self-citation, but not circularity in the load-bearing sense: the identity is independently verifiable by algebra, and it supports only the optional Condition II extension, not the main Proposition 1.

full rationale

The paper's central result, Proposition 1 (Ω_{Y|t} = Ω_{Y|do(t)} under positivity), is self-contained. The proof uses only the law of total probability, the backdoor adjustment formula, and the positivity assumption — all standard SCM machinery. No self-citation is involved. The only self-citation is to Sparkes et al. (2024) for the expected-average identity Av(Y) = E(Y) + κ and the U random variable class, which support the optional Condition II extension connecting Ω effects to mean contrasts. Even there, the identity is a mathematical fact derivable from the definition of κ, not an unverified ansatz or empirical claim. The self-citation is therefore real evidence (algebraically verifiable) rather than circular support. Proposition 2, which derives a formula for κ on discrete supports, is proven within the paper. No step in the main derivation chain reduces to its inputs by construction. Score 2 reflects the minor self-citation that is not load-bearing for the central claim.

Assumptions & free parameters 3 free parameters · 5 assumptions · 3 invented entities

The axiom ledger reveals a framework built on one standard assumption (positivity), two ad hoc conditions (II, III), and a mathematical identity from the author's own prior work. The invented entities (Ω estimands, κ, U random variables) form an interconnected conceptual system with no external validation.

free parameters (3)

Pr(θ) = 0.35 = 0.35
Subjective prior probability that treatment is effective, chosen by the author in the data application (Section 4). Not derived from any systematic elicitation.
Pr(y0,y*|θ) = 0.8 = 0.8
Subjective likelihood of observed data given treatment effectiveness, chosen by author (Section 4). No empirical basis provided.
Pr(y0,y*|θc) = 0.3 = 0.3
Subjective likelihood of observed data given treatment ineffectiveness, chosen by author (Section 4). No empirical basis provided.

assumptions (5)

domain assumption Positivity: f(t|u) > 0 for all t ∈ Ω_T and u ∈ Ω_U
Stated as Condition I in Section 3. The paper argues it holds when treatment is conceptualized coarsely and the DAG remains valid. It is the sole required condition for Proposition 1.
ad hoc to paper Epistemic preservation: κ_{Y|t} = κ_{Y|t'} for t ≠ t'
Stated as Condition II in Section 3. Requires that the agent assigns probability distributions with the same structural motif (same sum-symmetry/skew) across treatment values. No empirical validation provided.
ad hoc to paper Support recall: Ω_Ŷ|t=0 = Ω_Y|t=0
Stated as Condition III in Section 3. Requires that a client's recalled estimate of baseline outcomes draws on the same set of possible values as would have existed under measurement. Unverifiable.
standard math Expected-average identity: Av(Y) = E(Y) + κ
Invoked in Section 3 to connect support-based contrasts to mean contrasts. From Sparkes et al. (2024). A mathematical identity, not an empirical claim, but from the author's own prior work.
domain assumption de Finetti subjectivist interpretation of probability
Adopted throughout Section 2.1 and the entire paper. The proofs are stated to remain valid under other interpretations, but the practical meaning of the results changes radically.

invented entities (3)

Ω estimands
purpose: Class of causal estimands defined as contrasts of functions of outcome supports rather than distributions
New estimand class introduced by this paper. No external validation of its practical utility is provided; the paper acknowledges the open question of how often supports shift in practice.
κ (sum-symmetry/skew parameter)
purpose: Measures the distance between the functional average and expected value of a random variable; used in Condition II and the expected-average identity
Introduced in Sparkes et al. (2024). Not independently validated. Used to extend the framework from support contrasts to mean contrasts.
U random variables
purpose: Class of random variables with sum-symmetric distributions where Av(Y) = E(Y) and κ = 0
Introduced in Sparkes et al. (2024). Used to argue that Condition II is plausibly satisfied. Not independently validated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Causal Inference for Case Studies in Behavioral Health." pith.science (2026). https://pith.science/paper/SC5UBVGG

@misc{pith2026260706912,
  author       = {Pith},
  title        = {Pith review of: Causal Inference for Case Studies in Behavioral Health},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SC5UBVGG}},
  note         = {Machine review of arXiv:2607.06912}
}

abstract

We present a framework for causal inference in behavioral health case studies -- and observational N=1 settings more generally -- under unmeasured confounding. The framework rests on a class of causal estimands, termed $\Omega$ estimands, defined as contrasts of functions of an outcome variable's support rather than its distribution. Because such estimands do not depend on how probability is distributed over supports, they are insensitive to the confounding that limits other methods. We prove that, in a structural causal model, the observational and interventional supports of an outcome coincide under a single assumption -- positivity -- without any requirement that confounders be known, measured, or adjusted for. Two optional conditions extend the framework: one licensing a client's recalled baseline as a stand-in for sparsely measured baseline periods, and one connecting support contrasts to conventional mean contrasts through an expected-average identity. We adopt a subjectivist (de Finetti) interpretation of probability and situate the framework within mandates for measurement-based care. A case study of cognitive behavioral therapy for anxiety illustrates an elementary approach a provider can use.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages

[1]

2010 , publisher=

Causal inference , author=. 2010 , publisher=

work page 2010
[2]

Review of Economics and statistics , volume=

Nonparametric estimation of average treatment effects under exogeneity: A review , author=. Review of Economics and statistics , volume=. 2004 , publisher=

work page 2004
[3]

Epidemiology , volume=

The consistency statement in causal inference: a definition or an assumption? , author=. Epidemiology , volume=. 2009 , publisher=

work page 2009
[4]

Journal of Epidemiology & Community Health , volume=

Estimating causal effects from epidemiological data , author=. Journal of Epidemiology & Community Health , volume=. 2006 , publisher=

work page 2006
[5]

Epidemiology , volume=

Causal diagrams for epidemiologic research , author=. Epidemiology , volume=. 1999 , publisher=

work page 1999
[6]

Sociological Methodology , volume=

The foundations of causal inference , author=. Sociological Methodology , volume=. 2010 , publisher=

work page 2010
[7]

2009 , publisher=

Causality , author=. 2009 , publisher=

work page 2009
[8]

The international journal of biostatistics , volume=

An introduction to causal inference , author=. The international journal of biostatistics , volume=. 2010 , publisher=

work page 2010

Show all 45 references

[9]

Brown, Jonathan and Breslau, Joshua and Wishon, Allison and Miller, Rachel and Kase, Courtney and Dunbar, Michael and Stewart, Kate and Briscombe, Brian and Rose, Tyler and Dehus, Eric and DeWitt, Kathryn , title =
[10]

Implementation and Impacts of the Certified Community Behavioral Health Clinic Demonstration: Findings From the National Evaluation , author=
[11]

Psychiatric Services , pages=

Characterizing crisis services offered by certified community behavioral health clinics: results from a national survey , author=. Psychiatric Services , pages=. 2024 , publisher=

2024
[12]

Social Work in Mental Health , volume=

A brief report on certified community behavioral health clinics demonstration program , author=. Social Work in Mental Health , volume=. 2021 , publisher=

2021
[13]

2023 , url=

Certified Community Behavioral Health Center (CCBHC) Certification Criteria , author=. 2023 , url=

2023
[14]

2023 , eprint=

Properties and Deviations of Random Sums of Densely Dependent Random Variables , author=. 2023 , eprint=

2023
[15]

World Psychiatry , volume=

Etiology in psychiatry: embracing the reality of poly-gene-environmental causation of mental illness , author=. World Psychiatry , volume=. 2017 , publisher=

2017
[16]

2006 , publisher=

Psychiatric epidemiology: searching for the causes of mental disorders , author=. 2006 , publisher=

2006
[17]

World psychiatry , volume=

A network theory of mental disorders , author=. World psychiatry , volume=. 2017 , publisher=

2017
[18]

Administration and Policy in Mental Health and Mental Health Services Research , volume=

Individualized progress measures are more acceptable to clinicians than standardized measures: Results of a national survey , author=. Administration and Policy in Mental Health and Mental Health Services Research , volume=. 2018 , publisher=

2018
[19]

PSYCHLOPS

Measuring psychological outcomes after cognitive behaviour therapy in primary care: a comparison between a new patient-generated measure “PSYCHLOPS”(Psychological Outcome Profiles) and “HADS”(Hospital Anxiety and Depression Scale) , author=. Journal of Mental Health , volume=....

2009
[20]

2006 , publisher=

Foundations of Measurement: Additive and polynomial representations , author=. 2006 , publisher=

2006
[21]

, author=

Mathematical psychology: An elementary introduction. , author=. 1970 , publisher=

1970
[22]

Journal of Causal Inference , volume=

The functional average treatment effect , author=. Journal of Causal Inference , volume=. 2024 , publisher=

2024
[23]

Health psychology research , volume=

A cognitive-behavior therapy applied to a social anxiety disorder and a specific phobia, case study , author=. Health psychology research , volume=. 2014 , publisher=

2014
[24]

1974 , publisher =

Theory of Probability: A Critical Introductory Treatment , author =. 1974 , publisher =

1974
[25]

Holland , title =

Paul W. Holland , title =. Journal of the American Statistical Association , volume =. 1986 , publisher =

1986
[26]

American journal of physical medicine & rehabilitation , volume=

Case studies, single-subject research, and n of 1 randomized trials: Comparisons and Contrasts: 1 , author=. American journal of physical medicine & rehabilitation , volume=. 1999 , publisher=

1999
[27]

Health Psychology Review , volume=

Challenges and solutions for N-of-1 design studies in health psychology , author=. Health Psychology Review , volume=. 2019 , publisher=

2019
[28]

Genome medicine , volume=

Clinical trial design in the era of precision medicine , author=. Genome medicine , volume=. 2022 , publisher=

2022
[29]

2020 , publisher=

Single-case research designs: Methods for clinical and applied settings , author=. 2020 , publisher=

2020
[30]

Journal of Counseling & Development , volume=

The best kept secret in counseling: Single-case (N= 1) experimental designs , author=. Journal of Counseling & Development , volume=. 2000 , publisher=

2000
[31]

Remedial and Special Education , volume=

Single-case intervention research design standards , author=. Remedial and Special Education , volume=. 2013 , publisher=

2013
[32]

single-case Research Design and Analysis (psychology Revivals) , pages=

Single-case research design and analysis: An overview , author=. single-case Research Design and Analysis (psychology Revivals) , pages=. 2015 , publisher=

2015
[33]

Harvard data science review , volume=

Personalized data science and personalized (N-of-1) trials: Promising paradigms for individualized health care , author=. Harvard data science review , volume=. 2022 , doi=

2022
[34]

Harvard data science review , volume=

Conduct and implementation of personalized trials in research and practice , author=. Harvard data science review , volume=. 2022 , doi=

2022
[35]

Journal of clinical epidemiology , volume=

Single-patient (n-of-1) trials: a pragmatic clinical decision methodology for patient-centered comparative effectiveness research , author=. Journal of clinical epidemiology , volume=. 2013 , publisher=

2013
[36]

Current directions in psychological science , volume=

The new person-specific paradigm in psychology , author=. Current directions in psychological science , volume=. 2009 , publisher=

2009
[37]

Measurement , volume=

A manifesto on psychology as idiographic science: Bringing the person back into scientific psychology, this time forever , author=. Measurement , volume=. 2004 , publisher=

2004
[38]

Journal for Person-Oriented Research , volume=

On the relation between person-oriented and subject-specific approaches , author=. Journal for Person-Oriented Research , volume=. 2015 , doi=

2015
[39]

British Journal of Clinical Pharmacology , volume=

N-of-1 trials in clinical research: Methodological foundations, statistical approaches and implementation challenges , author=. British Journal of Clinical Pharmacology , volume=. 2026 , publisher=

2026
[40]

1962 , publisher =

How to Do Things with Words , author =. 1962 , publisher =

1962
[41]

1979 , publisher =

Expression and Meaning: Studies in the Theory of Speech Acts , author =. 1979 , publisher =

1979
[42]

Journal of general internal medicine , volume=

The PHQ-9: validity of a brief depression severity measure , author=. Journal of general internal medicine , volume=. 2001 , publisher=

2001
[43]

and Shim, Ruth S

Frank, Richard G. and Shim, Ruth S. , title =. Psychiatric Services , year =
[44]

, title =

Kjellsson, Gustav and Clarke, Philip and Gerdtham, Ulf-G. , title =. Journal of Health Economics , year =
[45]

and Slade, Tim and Hobbs, Megan J

Mewton, Louise and Kessler, Ronald C. and Slade, Tim and Hobbs, Megan J. and Brownhill, Louise and Birrell, Louise and Tonks, Zoe and Teesson, Maree and Newton, Nicola and Chapman, Cath and others , title =. Psychological Assessment , year =

Pith tools

Reviewed July 9, 2026 · model on record in the stance chip above.

[1] [1]

2010 , publisher=

Causal inference , author=. 2010 , publisher=

work page 2010

[2] [2]

Review of Economics and statistics , volume=

Nonparametric estimation of average treatment effects under exogeneity: A review , author=. Review of Economics and statistics , volume=. 2004 , publisher=

work page 2004

[3] [3]

Epidemiology , volume=

The consistency statement in causal inference: a definition or an assumption? , author=. Epidemiology , volume=. 2009 , publisher=

work page 2009

[4] [4]

Journal of Epidemiology & Community Health , volume=

Estimating causal effects from epidemiological data , author=. Journal of Epidemiology & Community Health , volume=. 2006 , publisher=

work page 2006

[5] [5]

Epidemiology , volume=

Causal diagrams for epidemiologic research , author=. Epidemiology , volume=. 1999 , publisher=

work page 1999

[6] [6]

Sociological Methodology , volume=

The foundations of causal inference , author=. Sociological Methodology , volume=. 2010 , publisher=

work page 2010

[7] [7]

2009 , publisher=

Causality , author=. 2009 , publisher=

work page 2009

[8] [8]

The international journal of biostatistics , volume=

An introduction to causal inference , author=. The international journal of biostatistics , volume=. 2010 , publisher=

work page 2010