The DOI matches its Crossref record; 30 of the 31 references it cites were found in a registry or an index; 1 of the 11 sources its claims cite were checked for support.
Linear Models: A Useful "Microscope" for Causal Analysis
Open accessAbstract
This note reviews basic techniques of linear path analysis and demonstrates, using simple examples, how causal phenomena of non-trivial character can be understood, exemplified and analyzed using diagrams and a few algebraic steps. The techniques allow for swift assessment of how various features of the model impact the phenomenon under investigation. This includes: Simpson’s paradox, case–control bias, selection bias, missing data, collider bias, reverse regression, bias amplification, near instruments, and measurement errors.
Overview
CoChat Summary
TL;DR
This note is a tutorial arguing that linear structural equation models and Wright's path-tracing rules serve as a "microscope" for causal analysis, letting researchers derive closed-form expressions for causal phenomena with elementary algebra and diagram inspection. It works through a catalogue of phenomena — Simpson's paradox, case–control and selection bias, missing data, collider bias, reverse regression, bias amplification, near instruments, and measurement error — showing how each depends on structural features of the model.
Key findings
- Path tracing plus the reduction of partial to pairwise regressions lets analysts read partial regression coefficients directly off a path diagram, and thereby assess how model features affect the phenomenon under study.
- Simpson's paradox arises when the confounding product has a different sign from the direct effect, so the regression of Y on X can change sign upon conditioning on Z for every value z; the magnitude of the change depends on the extent to which X and Y are confounded.
- Conditioning on a descendant of Y (case–control/selection bias) induces bias even though the variable is not on the pathway between X and Y, whereas conditioning on a descendant of X introduces no bias; in the general selection model, substantial bias can persist even when X and Y are uncorrelated.
- Bias amplification: conditioning on an instrumental variable produces a bias proportional to the pre-existing bias and increasing with the instrument's strength in predicting X — "the better Z predicts X, the higher the bias." For a near instrument that is also a confounder, Z reduces bias only if its effect on Y exceeds its effect on X by a stated factor.
- M-bias: adjusting for a pre-treatment collider Z introduces bias where none exists, and the bias increases substantially as Z becomes a good predictor of X — precisely when textbook reasons favour adjustment. The paper notes this refutes the folklore that a variable uncorrelated with both treatment and outcome can be exonerated from confounding considerations.
- Measurement error: adjusting for a proxy Z of an unobserved confounder U does not remove bias, but the bias vanishes as the proxy's reliability approaches unity; identical behaviour emerges when Z is a driver of U rather than a proxy.
- Missing data: the two missingness models considered are statistically indistinguishable yet each dictates a different estimation procedure, demonstrating that no model-blind estimator can guarantee an unbiased estimate even when one exists.
Methods
- Design: a methodological/tutorial note reviewing basic techniques of linear path analysis and demonstrating them on simple worked examples; no empirical data are analyzed.
- Core tools: standard definitions of variance, covariance, regression and correlation; recursive reduction of partial regression/correlation to pairwise coefficients; linear SEMs interpreted as assignment processes; and Wright's (1921) path-tracing rules, extended to read partial correlations from diagrams.
- Analysis: closed-form algebraic derivations of the regression coefficient under conditioning for each of eleven illustrative path diagrams, used to quantify bias and to generate counter-examples to over-ambitious conjectures.
Limitations
- Stated: the techniques are limited to linear analysis and hence restricted to homogeneous populations with no interactions; the linearity/effect-homogeneity assumption amounts to claiming all units react equally to any treatment, which "may exclude applications with profoundly heterogeneous subpopulations."
- Stated: conditional associations in this framework do not depend on the level z at which conditioning occurs, which makes linear analysis easy to manage but "limited in the spectrum of relationships it can capture."
- Stated: the paper positions the approach as superior to simulation studies only when conceptual understanding is of essence and problem size is manageable.
- Evident from the text: the derivations are confined to small, hand-tractable diagrams, and several results (e.g., the missing-data and measurement-error analyses) are asserted for the linear case with the claim that they hold nonparametrically, without proof in this note.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
1 Valid · 1 Not valid · 5 Not checked · 47 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Recent work has shown, however, that such adjustment would amplify the bias created by U [25–27].”
§ 3 The microscope at work: examples and their implications > 3.8 Bias amplification
What the sources say
- Partly supportedcites [25] Bhattacharya J, Vogt W. Do instrumental variables belong in propensity scores? Tech. Rep. NBER Technical Working Paper 343, National Bureau of Economic Research, MA, 2007.10.3386/t0343 sourceExperimental — accuracy not yet measured
Section 2.2 shows that regression-based propensity-score adjustment using an instrument multiplies the naive estimator’s bias by a factor greater than or equal to one. The claim omits this method-specific qualification: Section 2.3 says the differences in inconsistency cannot be signed in general for other propensity-score methods. (LLM, conf=0.86)
- Not matchedcites [26] Pearl J. On a class of bias-amplifying variables that endanger effect estimates. In Proceedings of the Twenty-Sixth Conference on Uncertainty in Artificial Intelligence. AUAI, Corvallis, OR, 417–24. http://ftp.cs.ucla.edu/pub/stat_ser/r356.pdf, 2010.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [27] Wooldridge J. Should instrumental variables be used as matching variables? Technical Report, https://www.msu.edu/ec/faculty/wooldridge/current%20research/treat1r6.pdf, Michigan State University, MI, 2009.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For Z to be a bias-reducer, its effect on Y (i.e., d) must exceed its effect on X (i.e., c) by a factor”
§ 3 The microscope at work: examples and their implications > 3.8 Bias amplification
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Adjusting for Z, which may be either pre-treatment or post-treatment covariate, introduces bias where none exists.”
§ 3 The microscope at work: examples and their implications > 3.6 The M-bias
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“which unveils a bias of size”
§ 3 The microscope at work: examples and their implications > 3.2 Conditioning on intermediaries and their proxies
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Conditioning on Z blocks the confounding path through”
§ 3 The microscope at work: examples and their implications > 3.10 The butterfly
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
24 more claims
Create a free account to see all 54 claims and every source verdict.
- It may state two separate claims as one. 20
- The quote only partly states this claim. 5
- Probably not a substantial claim. 10
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Note that the order of estimation is critical in the two MNAR examples considered and depends on the structure of the graph; no model-blind estimator exists for the MNAR class [32, 33].”
From The Deductive Approach to Causal Inference (2014) · cites this paper as [32]
Citations
Cites
Sample Selection Bias as a Specification Error
Heckman, J. J. (1979) · Econometrica · doi:10.2307/1912352 · Cited by 28,779 · cited 1×
“This bias reflects what economists called “selection bias” (Heckman, 1979) and epidemiologists “case-control bias” (Robins, 2001), which occurs when only patients for whom the outcome Y is evidenced (e.”
BackgroundStatistical Analysis With Missing Data
Lahiff, M., Little, R. J. A., & Rubin, D. B. (1989) · Journal of the American Statistical Association · doi:10.2307/2289883 · Cited by 17,265
Causality: Models, Reasoning, and Inference
Pearl, J. (2009) · doi:10.1017/cbo9780511803161 · Cited by 11,653
Self-citationRosenbaum, P. R. (2002) · Springer Series in Statistics · doi:10.1007/978-1-4757-3692-2 · Cited by 2,394
Causal diagrams for empirical research
PEARL, J. (1995) · Biometrika · doi:10.1093/biomet/82.4.669 · Cited by 2,355 · cited 3×
“[4]–[6], when combined with Wright’s path-tracing rules permits us to extend the latter so as to read partial correlations directly from the diagram.”
BackgroundSelf-citationMartuza, V. R., & Kenny, D. A. (1981) · Educational Researcher · doi:10.2307/1175080 · Cited by 2,028
Limitations of the Application of Fourfold Table Analysis to Hospital Data
Berkson, J. (1946) · Biometrics Bulletin · doi:10.2307/3002000 · Cited by 1,810 · cited 2×
“It has been known as Berkson Paradox [16], “explaining away” [17] or simply “collider bias.”
BackgroundKenny, D. A. (1979) · Medical Entomology and Zoology · Cited by 1,376
Introduction to Structural Equation Models.
Cherlin, A., & Duncan, O. D. (1976) · Contemporary Sociology · doi:10.2307/2064059 · Cited by 1,190
Hirano, K., & Imbens, G. W. (2001) · Health Services and Outcomes Research Methodology · doi:10.1023/a:1020371312283 · Cited by 1,039 · cited 1×
“This indistinguishability, together with the fact that Z may be a strong predictor of X may lure investigators to condition on Z to obtain an unbiased estimate of d [24].”
Background
Research works printed but not linked: 3
- Heise D. Causal analysis. New York: John Wiley and Sons, 1975.
- Crámer H. Mathematical methods of statistics. Princeton, NJ: Princeton University Press, 1946.
- Kim J, Pearl J. A computational model for combined causal and diagnostic reasoning in inference systems. In Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83). Karlsruhe, Germany, 1983.
Cited by
Collider bias undermines our understanding of COVID-19 disease risk and severity
Griffith, G. J., Morris, T. T., Tudball, M. J., Herbert, A., Mancano, G., Pike, L., Sharp, G. C., Sterne, J., Palmer, T. M., Davey Smith, G., et al. (2020) · Nature Communications · doi:10.1038/s41467-020-19478-2 · Cited by 826
A crash course in good and bad controls
Cinelli, C., Forney, A., & Pearl, J. (2024) · Sociological Methods & Research · doi:10.1177/00491241221099552 · Cited by 490
“The few exceptions that do discuss the problem of “ bad controls ” unfortunately cover only a narrow aspect of the problem (e.g. Angrist and Pischke, 2009, 2014; Wooldridge, 2010; Imbens and Rubin, 2015; Gelman, Hill and Vehtari, 2020).”
InfluentialSelf-citationInterpretation and identification of causal mediation
Pearl, J. (2014) · Psychological Methods · doi:10.1037/a0036434 · Cited by 554
“12 (See Bollen & Pearl, 2013, for further discussion of this important observation, which is often overlooked in the potential-outcome literature; e.g., Rubin, 2010; Sobel, 2008.)”
Self-citationSimpson's paradox in psychological science: a practical guide
Kievit, R. A., Frankenhuis, W. E., Waldorp, L. J., & Borsboom, D. (2013) · Frontiers in Psychology · doi:10.3389/fpsyg.2013.00513 · Cited by 542
“Simpson’s paradox is conceptually and analytically related to many statistical challenges and techniques, including causal inference (Pearl, 2000, 2013), the ecological fallacy (Robinson, 1950; Kramer, 1983; King, 1997; King and Roberts, 2012), Lord’s paradox, (Tu et al., 2008), propensity score…”
External validity: From do-calculus to transportability across populations
Pearl, J., & Bareinboim, E. (2013) · Cited by 398
Self-citation
Metadata
- Authors
- Pearl, J.
- Affiliations
- University of California, Los Angeles (US)
- Corresponding author
- Pearl, Judea
- Venue
- Journal of Causal Inference
- Publisher
- Walter de Gruyter GmbH
- Published
- 2013-05-29
- Volume
- 1
- Issue
- 1
- Pages
- 155-170
- ISSN
- 2193-3677 (ISSN-L)
- 2193-3685
- OpenAlex
- W2119047657
- Semantic Scholar
- 6358a3f7529ab731883119de82f2bbb2e115b11b
- Cited by
- 132
- References
- 31
- Citations by year
- 2026: 7 · 2025: 13 · 2024: 17 · 2023: 6 · 2022: 16 · 2021: 11 · 2020: 7 · 2019: 11 · 2018: 5 · 2017: 6 · 2016: 5 · 2015: 6 · 2014: 16 · 2013: 6
- FWCI
- 9.67
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Journal of Causal Inference (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- eScholarship (California Digital Library) (submittedVersion, cc-by)
- Funders
- National Science Foundation (1249822)
- Office of Naval Research (N00014)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Bayesian Modeling and Causal Inference
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Inference
- Keywords
- causal inference, Simpson's paradox, case-control bias, selection bias, missing data, collider bias, reverse regression, bias amplification, measurement error, causal graphs
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-30, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-09-30 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Linear Models: A Useful "Microscope" for Causal Analysis. https://app.cochat.ai/papers/10.1515/jci-2013-0003
- The paper
- Pearl, J. (2013). Linear Models: A Useful "Microscope" for Causal Analysis. Journal of Causal Inference. https://doi.org/10.1515/jci-2013-0003
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.