# Of Causation and Its Instruments: A Meditation on Knowledge and Authority **I. The Banishment and Its Pretense** Science has done what men in power do: mistaken constraint for virtue. For a century, causal language was expelled from respectable inquiry—not because causation ceased to operate in nature, but because its invocation could not be mechanically verified. The exile was called rigor. This is the familiar substitution: when we cannot master a thing, we declare mastery unnecessary. We renamed our limitation a principle. The statistician observed correlation. The physicist measured relations between quantities. Neither spoke of *why*. To ask why was to invite metaphysics, superstition, the ghost in the machine. Better to remain silent. Silence, it was reasoned, protects us from error. This reasoning is itself an error. Silence does not protect; it merely conceals the error we are already committing. **II. Pearl's Reversal: The Diagram as Prerequisite** Judea Pearl restored what should never have been exiled. He demonstrated that causal inference is not mysticism but mathematics—that we can reason rigorously *about* causation by first making our assumptions *explicit*. This is genuine rigor: not the avoidance of causal language, but its disciplined deployment. Pearl's instrument is the causal diagram. Before data speaks, a structure must be assumed. Before calculation, a map. This is not weakness but honesty. Every inference rests on prior commitments. Pearl simply insisted we name them. Yet here lies the trap, and it is a profound one: **the diagram is not discovered. It is imposed.** **III. The Transparency Deception** A researcher may draw her causal diagram with perfect clarity. Every arrow is justified. Every exclusion explained. The model is transparent—visible in all its parts, auditable, explicit. And it may be entirely wrong. Transparency and validity are not twins. They are sometimes strangers. A false model, fully articulated, remains false. The researcher who says "here is my assumption" has not thereby validated it. She has only made her error legible. This is dangerous because legibility breeds confidence. An explicit model appears more trustworthy than an implicit one. We confuse *knowing what we assume* with *knowing that our assumptions are correct*. The diagram becomes a kind of incantation: once drawn and displayed, it seems to consecrate the analysis that follows. **IV. Authority and the Diagram** Who decides what the diagram is? The answer reveals the hidden structure of causal inquiry. It is not data that decides. Data cannot speak until the diagram permits it. It is not nature—nature is mute about its own architecture. The diagram is decided by the researcher, the domain expert, the consensus of a field, the funding agency, the prevailing theory. In short: it is decided by *power*. A causal diagram is an assertion of authority. It says: this is how the world hangs together. And because it is explicit, it appears democratic—open to challenge, subject to scrutiny. But the challenge is bounded. You may dispute the diagram, but only if you propose another. You cannot proceed without one. The framework itself is not negotiable; only the contents are. Consider: two researchers, equally rigorous, may draw contradictory diagrams. Both are transparent. Both are mathematically sound. Both are *provably correct* given their respective diagrams. Yet they cannot both be right about the world. One diagram must be wrong. But the mathematics cannot tell you which. The mathematics only tells you what follows *if* you accept the diagram's truth. **V. The Causal Dimension: What It Means to Know a Cause** To know the cause of something is not merely to identify a prior event that is reliably followed by another. A cause is not a correlation extended backward in time. To know a cause is to know what would happen if we *intervened*—if we reached into the causal chain and altered one element. The cause is the element whose alteration would produce change in the effect. This is Pearl's great insight: causation is defined by *counterfactual dependence*, not temporal sequence. But notice what this requires: we must know what the world would be like under conditions we have not observed. We must reason about possibilities, not facts. We must assume a structure that maps observed reality onto unobserved possibilities. This is where the diagram becomes indispensable—and where its arbitrariness becomes unbearable. **VI. The Problem of Multiple Realities** Consider a simple case: Does poverty cause crime? The causal diagram must specify whether poverty causes crime directly, or whether some third factor (lack of education, social disintegration, police surveillance) is the true cause, with poverty merely a symptom or correlate. The diagram must specify whether crime influences poverty—whether criminal records reduce economic opportunity. It must specify what is *exogenous* (caused by factors outside the system) and what is *endogenous* (caused within it). Each choice produces a different causal model. Each model is internally consistent. Each, given its diagram, is mathematically sound. But they generate different answers to the question: "Does poverty cause crime?" Which diagram is correct? The answer cannot come from data alone. It comes from theory, from prior studies, from the researcher's judgment about how the world works. It comes, ultimately, from *authority*. The researcher with institutional power—the prestigious researcher, the one whose previous work is accepted, the one whose diagram aligns with prevailing opinion—will have her diagram adopted. Her causal claims will be treated as established. A challenger's diagram, equally rigorous, equally transparent, will be treated as speculative. **VII. The Dimension of Causal: A Reframing** We should not ask: "Is the diagram correct?" We should ask: "Whose purposes does this diagram serve?" A diagram that treats poverty as a cause of crime invites intervention at the level of poverty. It suggests policy: redistribute resources, create opportunity. A diagram that treats poverty and crime as co-symptoms of a deeper social disorder invites different policy: reform institutions, restore community. A diagram that treats crime as a cause of poverty invites yet another response: strengthen law enforcement, expand incarceration. The causal diagram is not a neutral instrument. It is a tool of governance. It determines what interventions seem rational, what problems seem solvable, what actors bear responsibility. This is not an argument against causal reasoning. It is an argument for honesty about what causal reasoning is: an exercise of power dressed in the language of mathematics. **VIII. Conclusion: The Necessity of Judgment** We cannot escape the need for causal diagrams. Data does not speak without them. Inference is impossible without prior structure. Pearl was right to restore causal language, to make it rigorous, to insist that we be explicit about our assumptions. But we must also be explicit about this: *the diagram is a choice, not a discovery*. It reflects judgment, interest, authority. Its transparency does not validate it. Its mathematical consistency does not make it true. The researcher who understands this—who knows that her diagram is an imposition, not a revelation—may proceed with appropriate humility. She will hold her causal claims lightly. She will attend to alternative diagrams. She will remember that her conclusions depend crucially on commitments that data cannot verify. This is the only rigor available to us: not the false rigor of avoiding causal language, and not the false rigor of drawing causal diagrams and forgetting they are drawn. It is the difficult rigor of knowing what we assume, why we assume it, and what alternatives we have rejected—and being willing to say so.