Causal inference is the process of determining the independent, actual effect of a particular phenomenon that is a component of a larger system. The main difference between causal inference and inference of association is that causal inference analyzes the response of an effect variable when a cause of the effect variable is changed.
The study of why things occur is called etiology, and can be described using the language of scientific causal notation. Causal inference is said to provide the evidence of causality theorized by causal reasoning. Causal inference is widely studied across all sciences. Several innovations in the development and implementation of methodology designed to determine causality have proliferated in recent decades. Causal inference remains especially difficult where experimentation is difficult or impossible, which is common throughout most sciences.
The approaches to causal inference are broadly applicable across all types of scientific disciplines, and many methods of causal inference that were designed for certain disciplines have found use in other disciplines. This article outlines the basic process behind causal inference and details some of the more conventional tests used across different disciplines; however, this should not be mistaken as a suggestion that these methods apply only to those disciplines, merely that they are the most commonly used in that discipline.
Causal inference is difficult to perform and there is significant debate amongst scientists about the proper way to determine causality. Despite other innovations, there remain concerns of misattribution by scientists of correlative results as causal, of the usage of incorrect methodologies by scientists, and of deliberate manipulation by scientists of analytical results in order to obtain statistically significant estimates. Particular concern is raised in the use of regression models, especially linear regression models.
Definition Inferring the cause of something has been described as: "...reason[ing] to the conclusion that something is, or is likely to be, the cause of something else". "Identification of the cause or causes of a phenomenon, by establishing covariation of cause and effect, a time-order relationship with the cause preceding the effect, and the elimination of plausible alternative causes."
Methodology General
Causal inference is conducted via the study of systems where the measure of one variable is suspected to affect the measure of another. Causal inference is conducted with regard to the scientific method. The first step of causal inference is to formulate a falsifiable null hypothesis, which is subsequently tested with statistical methods. Frequentist statistical inference is the use of statistical methods to determine the probability that the data occur under the null hypothesis by chance; Bayesian inference is used to determine the effect of an independent variable. Statistical inference is generally used to determine the difference between variations in the original data that are random variation or the effect of a well-specified causal mechanism. Notably, correlation does not imply causation, so the study of causality is as concerned with the study of potential causal mechanisms as it is with variation amongst the data. A frequently sought after standard of causal inference is an experiment wherein treatment is randomly assigned but all other confounding factors are held constant. Most of the efforts in causal inference are in the attempt to replicate experimental conditions.
Epidemiological studies employ different epidemiological methods of collecting and measuring evidence of risk factors and effect and different ways of measuring association between the two. Results of a 2020 review of methods for causal inference found that using existing literature for clinical training programs can be challenging. This is because published articles often assume an advanced technical background, they may be written from multiple statistical, epidemiological, computer science, or philosophical perspectives, methodological approaches continue to expand rapidly, and many aspects of causal inference receive limited coverage. Common frameworks for causal inference include the causal pie model (component-cause), Pearl's structural causal model (causal diagram + do-calculus), structural equation modeling, and Rubin causal model (potential-outcome), which are often used in areas such as social sciences and epidemiology.
Experimental
Experimental verification of causal mechanisms is possible using experimental methods. The main motivation behind an experiment is to hold other experimental variables constant while purposefully manipulating the variable of interest. If the experiment produces statistically significant effects as a result of only the treatment variable being manipulated, there is grounds to believe that a causal effect can be assigned to the treatment variable, assuming that other standards for experimental design have been met.
Quasi-experimental
Quasi-experimental verification of causal mechanisms is conducted when traditional experimental methods are unavailable. This may be the result of prohibitive costs of conducting an experiment, or the inherent infeasibility of conducting an experiment, especially experiments that are concerned with large systems such as economies of electoral systems, or for treatments that are considered to present a danger to the well-being of test subjects. Quasi-experiments may also occur where information is withheld for legal reasons.
Approaches in epidemiology Epidemiology studies patterns of health and disease in defined populations of living beings in order to infer causes and effects. An association between an exposure to a putative risk factor and a disease may be suggestive of, but is not equivalent to causality because correlation does not imply causation. Historically, Koch's postulates have been used since the 19th century to decide if a microorganism was the cause of a disease. In the 20th century the Bradford Hill criteria, described in 1965 have been used to assess causality of variables outside microbiology, although even these criteria are not exclusive ways to determine causality. In molecular epidemiology the phenomena studied are on a molecular biology level, including genetics, where biomarkers are evidence of cause or effects.
A recent trend is to identify evidence for influence of the exposure on molecular pathology within diseased tissue or cells, in the emerging interdisciplinary field of molecular pathological epidemiology (MPE). Linking the exposure to molecular pathologic signatures of the disease can help to assess causality. Considering the inherent nature of heterogeneity of a given disease, the unique disease principle, disease phenotyping and subtyping are trends in biomedical and public health sciences, exemplified as personalized medicine and precision medicine.
Causal Inference has also been used for treatment effect estimation. Assuming a set of observable patient symptoms(X) caused by a set of hidden causes(Z) we can choose to give or not a treatment t. The result of the giving or not giving the treatment is the effect estimation y. If the treatment is not guaranteed to have a positive effect then the decision whether the treatment should be applied or not depends firstly on expert knowledge that encompasses the causal connections. For novel diseases, this expert knowledge may not be available. As a result, we rely solely on past treatment outcomes to make decisions. A modified variational autoencoder can be used to model the causal graph described above. While the above scenario could be modelled without the use of the hidden confounder(Z) we would lose the insight that the symptoms a patient together with other factors impacts both the treatment assignment and the outcome.
Approaches in computer science Causal inference is an important concept in the field of causal artificial intelligence. Determination of cause and effect from joint observational data for two time-independent variables, say X and Y, has been tackled using asymmetry between evidence for some model in the directions, X → Y and Y → X. The primary approaches are based on Algorithmic information theory models and noise models.
Noise models Incorporate an independent noise term in the model to compare the evidences of the two directions.
Here are some of the noise models for the hypothesis Y → X with the noise E: Additive noise:Hoyer, Patrik O., et al. "Nonlinear causal discovery with additive noise models ." NIPS. Vol. 21. 2008. Linear noise: Post-nonlinear:Zhang, Kun, and Aapo Hyvärinen. "On the identifiability of the post-nonlinear causal model ." Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence. AUAI Press, 2009. Heteroskedastic noise: Functional noise:Mooij, Joris M., et al. "Probabilistic latent variable models for distinguishing between cause and effect ." NIPS. 2010.
The common assumption in these models are: There are no other causes of Y. X and E have no common causes. Distribution of cause is independent from causal mechanisms.
On an intuitive level, the idea is that the factorization of the joint distribution P(Cause, Effect) into P(Cause)*P(Effect | Cause) typically yields models of lower total complexity than the factorization into P(Effect)*P(Cause | Effect). Although the notion of "complexity" is intuitively appealing, it is not obvious how it should be precisely defined. A different family of methods attempt to discover causal "footprints" from large amounts of labeled data, and allow the prediction of more flexible causal relations.Lopez-Paz, David, et al. "Towards a learning theory of cause-effect inference " ICML. 2015
Approaches in social sciences
Social science The social sciences in general have moved increasingly toward including quantitative frameworks for assessing causality. Much of this has been described as a means of providing greater rigor to social science methodology. Political science was significantly influenced by the publication of Designing Social Inquiry, by Gary King, Robert Keohane, and Sidney Verba, in 1994. King, Keohane, and Verba recommend that researchers apply both quantitative and qualitative methods and adopt the language of statistical inference to be clearer about their subjects of interest and units of analysis. Proponents of quantitative methods have also increasingly adopted the potential outcomes framework, developed by Donald Rubin, as a standard for inferring causality.
While much of the emphasis remains on statistical inference in the potential outcomes framework, social science methodologists have developed new tools to conduct causal inference with both qualitative and quantitative methods, sometimes called a "mixed methods" approach. Advocates of diverse methodological approaches argue that different methodologies are better suited to different subjects of study. Sociologist Herbert Smith and political scientists James Mahoney and Gary Goertz have cited the observation of Paul W. Holland, a statistician and author of the 1986 article "Statistics and Causal Inference", that statistical inference is most appropriate for assessing the "effects of causes" rather than the "causes of effects".