CAUSALITY, CAUSES, AND CAUSAL INFERENCE

Causality describes ideas about the nature of the relations of cause and effect. A cause is something that produces or occasions an effect. Causal inference is the thought process that tests whether a relationship of cause to effect exists.

CAUSALITY

Ideas about cause vary as culture and philosophical concepts generate their own momentum and logical requirements. In epidemiology, ideas of causality changed as societies, understanding of disease, and technical resources changed. In nineteenth-century England, William Farr, John Snow, John Simon, and others founded an ecologic epidemiology, imbued with the Sanitary Movement's concern about the wretched context of urbanization and industrialization and its attendant miasma theory.

Late in the century the advances of Louis Pasteur, Robert Koch, and others studying bacteria displaced the theory that miasma caused disease. In Germany, this specific-cause theory was challenged early in the twentieth century. Germ theory sought proximate, singular microbial causes of disease. For seventy-five years, however, it dominated an anglophone laboratory-based epidemiology and focused on infectious disease control.

By the mid–twentieth century, epidemic infectious disease was receding in the face of effective action and social change. Epidemic chronic diseases—cancer and respiratory or cardiac disorders—overrode other epidemiological concerns. A. Bradford Hill, Richard Doll, Jeremy Morris, and others led a search for causal factors in the new epidemics. The need became apparent for theory to accommodate multiple coexisting causes, now understood as risk factors.

Such shifts are inevitable for and epidemiology obliged, in a dynamic world, always to evolve new means and theory for coping. Unamended multiple cause theory too is unlikely to be enough to meet rising challenges. Despite its core in population level studies, epidemiology in the past half century concentrated on individual-level disease risk, but studies on several levels, including change over time, are increasingly frequent, however.

CAUSES

For contemporary epidemiologists, what elements constitute the idea of cause? This public health science is chiefly observational. Neither humans nor societies are readily manipulable as rigorous experiment requires, excepting circumstances such as permit and clinical trials, which can randomize groups to neutralize differences other than the experimental intervention.

In the observational sciences, inference about causes is well served by a pragmatic concept of determinants, here defined as something making a difference to outcome. Everyone would not accept so broad a sweep. Galileo's seventeenth-century formulation was that causes be "necessary and sufficient." In the experimental sciences this came to mean one to one relationships of specific causes to given effects. In 1840, Friedrich Henle developed a similar mode for the testing of his idea that infection, not miasma, was a cause for disease.

Later, Koch revised Henle in devising a set of postulates. These sealed Koch's work on bacteria as the proximate cause of tuberculosis. In the germ theory era, the postulates served to guide the search for specific organisms as one to one causes of given diseases. The postulates require laboratory-based experiment; they entail not only observation (does a singular bacterial distribution parallel disease-affected sites?) but intervention (do bacteria, cultured and then transmitted to animals, produce lesions analogous to human disease?).

Causality, thus defined by experiment with specified active agents producing change, excludes from consideration two other types. First, it excludes steady-state conditions, like sex or social position or climate or location. Being contextual, if not always passive, these cannot be actively structured to produce change, as is done in experiment. In the lexicon of determinants for noninterventionist science, such conditions are nonetheless determinants. Experiment cannot reproduce them; they are detected in controlled comparisons of different times and places.

Second, change upon rigorous experimental intervention cannot but exclude a panoply of determinants antecedent and subsequent to the change agent. A table of the possible combinations inherent in Galileo's causal requirements results in the following:

Here, determinants emerge as necessary and sufficient (row 1: both always present); necessary but not sufficient (row 2: always present but only with others); sufficient but not necessary (row 3: sometimes effective alone); and, probably most often, neither necessary nor sufficient but contributory (row 4). Thus, taken separately, Galileo's criteria encompass a multicausal theory that accommodates conditions and causal chains. Without multicausality, in truth, no theoretical legitimation exists for the chronic disease epidemiology of the twentieth century.

CAUSAL INFERENCE

The nature of cause must be agreed before it can be inferred. David Hume, an eighteenth-century Scottish philosopher, isolated three properties essential to cause. Freely translated, these are association (cause and effect occur together), time order (causes precede effects), and connection or direction (repeated demonstrable, hence predictable, linkages exist between cause and effect). Hume's analysis endures and has hardly been improved on.

Causal inference is a judgmental process, not a snapshot but a movie. Causal hypotheses arise either from observation, existing knowledge and inductive reason, or from intuition. Hypotheses enunciated beforehand (a priori) can be subjected to repeated tests that allow their elimination or survival in respect, successively, of the causal properties. For each, criteria (or guidelines, canons, or postulates) assist judgment. They can be grouped under five categories, some with subcategories: strength; specificity; consistency; predictive performance; and coherence. Each is useful depending on the property under test and the type and quality of available research evidence.

Association is judged by the presence and strength of probabilities based on preset expectations of variation (so-called chance occurrence) and by consistency upon replication. Given survival, tests for time-order rely on establishing the sequence of cause and effect; reversal assures elimination. Failing elimination, the acid test for the property of connection or direction is difficult indeed: It depends on the complete array of criteria, with all alternative explanations and confounding accounted for.

In counterpoint, Hume rejected the validity of inductive logic (generalization from assembled particular observations) and thereby created an enduring problem for causal inference. He could find, he said, no logical compulsion to believe the sun would rise tomorrow. For him, logic could not demonstrate "necessary connection" from cause to effect. Following Hume, Karl Popper in the twentieth century found proof of hypotheses (verification) beyond logical reach. His theory allowed conclusive rejection (falsification) alone. His epidemiological followers prosecuted an intense debate in support.

Popper, aiming to falsify hypotheses, insisted solely on deductive logic (prediction of particular outcomes from prior general hypotheses). He dismissed such longstanding counters to his view of science as Bacon's (1599) or John Stuart Mill's (1856) inductive logic. Mill spelled out inductive "canons" implicit in the casual inferences of much observational science, including epidemiology. Despite Popper's insistent rejection, all scientists practice induction. Obligatory pragmatists, many also recognize added value in Popper's "hypothetico-deductive" procedure.

Other perspectives exist in epidemiology. Kenneth Rothman espouses a Popper-like view of causal inference but also provides heuristic if nondynamic model of multiple "sufficient and component" causes. James Robins, Sander Greenland, and others, following one of Hume's ideas modeled by Jerzy Neyman in 1923, elaborate the "counterfactual" approach. Limiting causes to change agents, this excludes steady-state conditions as causative if not as context, and demands strict formulations amendable to mathematical logic. A thought experiment compares an entity after exposure to the same entity had there been no exposure, a comparison unattainable in practice. Instead, the outcome variable is adjusted statistically. Bayesian probability theory provides artillery for applications. Tests for counterfactuals in securing or dismissing uncertain causality in epidemiology are awaited with interest.

Mervyn W. Susser

(see also: Epidemiology )

Bibliography

Bacon, F. (1599). The New Organon and Related Writings. Reprint. New York: Bobbs-Merril, 1960.

Evans, A. S. (1993). Causation and Disease: A Chronological Journey. New York, London: Plenum.

Hill, A. B. (1965). "Environment and Disease: Association or Causation?" Proceedings of the Royal Society of Medicine 58:295–300.

Hume, D. (1739). A Treatise of Human Nature. Book 1, Of the Understanding. Section 2. Reprint. La Salle, IL: Open Court, 1945.

Koch, R. (1882). Dei aetiologie der Tuberkulose. Reprint. In Gesammelte Werke von Koch, ed. J. Schwalbe. Leipzig: Georg Thieme, 1912.

MacMahon, B.; Pugh, T.; and Ipsen, J. (1960). Principles of Epidemiology. Boston: Little Brown.

Mill, J. S. (1856). A System of Logic: Ratiocinative and Inductive. Reprint. London: Rutledge, 1892.

Neyman, J. (1923). "On the Application of Probability Theory to Agricultural Experiments." Translation of excerpts by D. Dabrowska and T. Speed. Statistical Science 5 (1990):462–472.

Pearl, J. (2000). Causality. New York: Cambridge University Press.

Popper, K. R. (1968). The Logic of Scientific Discovery, revised edition. New York: Harper and Row. Originally published as Logik der Forschung (Vienna: Springer, 1934).

Rosen, G. (1941). "Jacob Henle and William Farr." Bulletin of History and Medicine 9:585–589.

—— (1958). A History of Public Health. New York: M.D. Publications.

Rothman, K. J., ed. (1988). Causal Inference. Chestnut Hill, MA: Epidemiology Resources.

Susser, M. (1991). "What Is a Cause and How Do We Know One? A Grammar for Pragmatic Epidemiology." American Journal of Epidemiology 33:635–648.

Van Engelhard, D. (1996). "Causality and Conditionality in Medicine Around 1900." In Science, Technology, and the Art of Medicine, eds. C. Delkeskamp-Hayes and M. A. G. Cutter. The Netherlands: Kluwer Adademic Publishers.