Causal graph

In statistics, econometrics, epidemiology, genetics and related disciplines, causal graphs (also known as path diagrams, causal Bayesian networks or DAGs) are probabilistic graphical models used to encode assumptions about the data-generating process.

Causal graphs can be used for communication and for inference. They are complementary to other forms of causal reasoning, for instance using causal equality notation. As communication devices, the graphs provide formal and transparent representation of the causal assumptions that researchers may wish to convey and defend. As inference tools, the graphs enable researchers to estimate effect sizes from non-experimental data, ^{[1] [2] [3] [4] [5]} derive testable implications of the assumptions encoded, ^{[1] [6] [7] [8]} test for external validity, ^[9] and manage missing data ^[10] and selection bias. ^[11]

Causal graphs were first used by the geneticist Sewall Wright ^[12] under the rubric "path diagrams". They were later adopted by social scientists ^{[13] [14] [15] [16] [17]} and, to a lesser extent, by economists. ^[18] These models were initially confined to linear equations with fixed parameters. Modern developments have extended graphical models to non-parametric analysis, and thus achieved a generality and flexibility that has transformed causal analysis in computer science, epidemiology, ^[19] and social science. ^[20] Recent advances include the development of large-scale causality graphs, such as CauseNet, which compiles over 11 million causal relations extracted from web sources to support causal question answering and reasoning. ^[21]

Construction and terminology

The causal graph can be drawn in the following way. Each variable in the model has a corresponding vertex or node and an arrow is drawn from a variable X to a variable Y whenever Y is judged to respond to changes in X when all other variables are being held constant. Variables connected to Y through direct arrows are called parents of Y, or "direct causes of Y," and are denoted by Pa(Y).

Causal models often include "error terms" or "omitted factors" which represent all unmeasured factors that influence a variable Y when Pa(Y) are held constant. In most cases, error terms are excluded from the graph. However, if the graph author suspects that the error terms of any two variables are dependent (e.g. the two variables have an unobserved or latent common cause) then a bidirected arc is drawn between them. Thus, the presence of latent variables is taken into account through the correlations they induce between the error terms, as represented by bidirected arcs.

Fundamental tools

A fundamental tool in graphical analysis is d-separation, which allows researchers to determine, by inspection, whether the causal structure implies that two sets of variables are independent given a third set. In recursive models without correlated error terms (sometimes called Markovian), these conditional independences represent all of the model's testable implications. ^[22]

Example

Suppose we wish to estimate the effect of attending an elite college on future earnings. Simply regressing earnings on college rating will not give an unbiased estimate of the target effect because elite colleges are highly selective, and students attending them are likely to have qualifications for high-earning jobs prior to attending the school. Assuming that the causal relationships are linear, this background knowledge can be expressed in the following structural equation model (SEM) specification.

Model 1

${\displaystyle {\begin{aligned}Q_{1}&=U_{1}\\C&=a\cdot Q_{1}+U_{2}\\Q_{2}&=c\cdot C+d\cdot Q_{1}+U_{3}\\S&=b\cdot C+e\cdot Q_{2}+U_{4},\end{aligned}}}$

where ${\displaystyle Q_{1}}$ represents the individual's qualifications prior to college, ${\displaystyle Q_{2}}$ represents qualifications after college, ${\displaystyle C}$ contains attributes representing the quality of the college attended, and ${\displaystyle S}$ the individual's salary.

Figure 1: Unidentified model with latent variables (${\displaystyle Q_{1}}$ and ${\displaystyle Q_{2}}$) shown explicitly

Figure 2: Unidentified model with latent variables summarized

Figure 1 is a causal graph that represents this model specification. Each variable in the model has a corresponding node or vertex in the graph. Additionally, for each equation, arrows are drawn from the independent variables to the dependent variables. These arrows reflect the direction of causation. In some cases, we may label the arrow with its corresponding structural coefficient as in Figure 1.

If ${\displaystyle Q_{1}}$ and ${\displaystyle Q_{2}}$ are unobserved or latent variables their influence on ${\displaystyle C}$ and ${\displaystyle S}$ can be attributed to their error terms. By removing them, we obtain the following model specification:

Model 2

${\displaystyle {\begin{aligned}C&=U_{C}\\S&=\beta C+U_{S}\end{aligned}}}$

The background information specified by Model 1 imply that the error term of ${\displaystyle S}$, ${\displaystyle U_{S}}$, is correlated with C's error term, ${\displaystyle U_{C}}$. As a result, we add a bidirected arc between S and C, as in Figure 2.

Figure 3: Identified model with latent variables (${\displaystyle Q_{1}}$ and ${\displaystyle Q_{2}}$) shown explicitly

Figure 4: Identified model with latent variables summarized

Since ${\displaystyle U_{S}}$ is correlated with ${\displaystyle U_{C}}$ and, therefore, ${\displaystyle C}$, ${\displaystyle C}$ is endogenous and ${\displaystyle \beta }$ is not identified in Model 2. However, if we include the strength of an individual's college application, ${\displaystyle A}$, as shown in Figure 3, we obtain the following model:

Model 3

${\displaystyle {\begin{aligned}Q_{1}&=U_{1}\\A&=a\cdot Q_{1}+U_{2}\\C&=b\cdot A+U_{3}\\Q_{2}&=e\cdot Q_{1}+d\cdot C+U_{4}\\S&=c\cdot C+f\cdot Q_{2}+U_{5},\end{aligned}}}$

By removing the latent variables from the model specification we obtain:

Model 4

${\displaystyle {\begin{aligned}A&=a\cdot Q_{1}+U_{A}\\C&=b\cdot A+U_{C}\\S&=\beta \cdot C+U_{S},\end{aligned}}}$

with ${\displaystyle U_{A}}$ correlated with ${\displaystyle U_{S}}$.

Now, ${\displaystyle \beta }$ is identified and can be estimated using the regression of ${\displaystyle S}$ on ${\displaystyle C}$ and ${\displaystyle A}$. This can be verified using the single-door criterion, ^{[1] [23]} a necessary and sufficient graphical condition for the identification of a structural coefficients, like ${\displaystyle \beta }$, using regression.

References

1. Pearl, Judea (2000). Causality . Cambridge, MA: MIT Press. ISBN   9780521773621.
2. Tian, Jin; Pearl, Judea (2002). "A general identification condition for causal effects". Proceedings of the Eighteenth National Conference on Artificial Intelligence. ISBN   978-0-262-51129-2.
3. Shpitser, Ilya; Pearl, Judea (2008). "Complete Identification Methods for the Causal Hierarchy" (PDF). Journal of Machine Learning Research. 9: 1941–1979.
4. Huang, Y.; Valtorta, M. (2006). Identifiability in Causal Bayesian Networks: A Sound and Complete Algorithm (PDF).
5. Bareinboim, Elias; Pearl, Judea (2012). "Causal Inference by Surrogate Experiments: z-Identifiability". Proceedings of the Twenty-Eighth Conference on Uncertainty in Artificial Intelligence. arXiv: 1210.4842 . Bibcode:2012arXiv1210.4842B. ISBN   978-0-9749039-8-9.
6. Tian, Jin; Pearl, Judea (2002). "On the Testable Implications of Causal Models with Hidden Variables". Proceedings of the Eighteenth Conference on Uncertainty in Artificial Intelligence. pp. 519–27. arXiv: 1301.0608 . Bibcode:2013arXiv1301.0608T. ISBN   978-1-55860-897-9.
7. Shpitser, Ilya; Pearl, Judea (2008). "Complete Identification Methods for the Causal Hierarchy" (PDF). Journal of Machine Learning Research. 9 (64): 1941–1979. ISSN   1533-7928 . Retrieved 2024-08-11.
8. Chen, Bryant; Pearl, Judea (2014). "Testable Implications of Linear Structural Equation Models". Proceedings of the AAAI Conference on Artificial Intelligence. 28. doi: 10.1609/aaai.v28i1.9065 . S2CID   1612893.
9. Bareinmboim, Elias; Pearl, Judea (2014). "External Validity: From do-calculus to Transportability across Populations". Statistical Science. 29 (4): 579–595. arXiv: 1503.01603 . doi:10.1214/14-sts486. S2CID   5586184.
10. Mohan, Karthika; Pearl, Judea; Tian, Jin (2013). "Graphical Models for Inference with Missing Data" (PDF). Advances in Neural Information Processing Systems.
11. Bareinboim, Elias; Tian, Jin; Pearl, Judea (2014). "Recovering from Selection Bias in Causal and Statistical Inference". Proceedings of the AAAI Conference on Artificial Intelligence. 28. doi: 10.1609/aaai.v28i1.9074 .
12. Wright, S. (1921). "Correlation and causation". Journal of Agricultural Research. 20: 557–585.
13. Blalock, H. M. (1960). "Correlational analysis and causal inferences". American Anthropologist. 62 (4): 624–631. doi: 10.1525/aa.1960.62.4.02a00060 .
14. Duncan, O. D. (1966). "Path analysis: Sociological examples". American Journal of Sociology. 72: 1–16. doi:10.1086/224256. S2CID   59428866.
15. Duncan, O. D. (1976). "Introduction to structural equation models". American Journal of Sociology. 82 (3): 731–733. doi:10.1086/226377.
16. Jöreskog, K. G. (1969). "A general approach to confirmatory maximum likelihood factor analysis". Psychometrika. 34 (2): 183–202. doi:10.1007/bf02289343. S2CID   186236320.
17. Goldberger, A. S. (1972). "Structural equation models in the social sciences". Econometrica. 40 (6): 979–1001. doi:10.2307/1913851. JSTOR   1913851.
18. White, Halbert; Chalak, Karim; Lu, Xun (2011). "Linking granger causality and the pearl causal model with settable systems" (PDF). Causality in Time Series Challenges in Machine Learning. 5.
19. Rothman, Kenneth J.; Greenland, Sander; Lash, Timothy (2008). Modern epidemiology. Lippincott Williams & Wilkins. ISBN   978-0-7817-5564-1.
20. Morgan, S. L.; Winship, C. (2007). Counterfactuals and causal inference: Methods and principles for social research. New York: Cambridge University Press. doi:10.1017/cbo9781107587991. ISBN   978-1-107-06507-9.
21. Heindorf, Stefan; Scholten, Yan; Wachsmuth, Henning; Ngonga Ngomo, Axel-Cyrille; Potthast, Martin (2020). "CauseNet: Towards a Causality Graph Extracted from the Web". Proceedings of the 29th ACM International Conference on Information & Knowledge Management. CIKM. ACM.
22. Geiger, Dan; Pearl, Judea (1993). "Logical and Algorithmic Properties of Conditional Independence". Annals of Statistics. 21 (4): 2001–2021. CiteSeerX   10.1.1.295.2043 . doi:10.1214/aos/1176349407.
23. Chen, B.; Pearl, J (2014). "Graphical Tools for Linear Structural Equation Modeling" (PDF). Technical Report.