Axiomatic causal calculus converts do-operand-based probability formulations into conditional probability notations. The graphical criteria for when specific substitutions are allowed are provided using three axiom schemas.

The causal calculus employs Bayesian conditioning, p(y|x), in which x is the observed variable, and causal conditioning, p(y|do(x)), in which an action is made to compel a given value x.

The inference of a cause-and-effect relationship from observational studies must be based on some qualitative theoretical assumptions when experimental interventions are impractical or forbidden, such as the idea that symptoms do not cause diseases. These assumptions are typically expressed as missing arrows in causal graphs like Bayesian networks or path diagrams. 

Causal calculus theory

Differentiating between conditional probabilities, such as P(cancer|smoking), and interventional probabilities, such as P(cancer|do(smoking)), is central to the theory behind these derivations. The first states, "the probability of finding cancer in a person known to smoke, having begun, unforced by the experimenter, to do so at some unspecified time in the past," while the second states, "the probability of finding cancer in a person forced by the experimenter to smoke at some specified time in the past." 

The former is a statistical concept that may be estimated through observation with minimal experimenter intervention. In contrast, the latter is a causal concept that must be assessed through a controlled and randomly administered experiment. The observer effect provides a quantitative description of a defining feature of quantum phenomena: the necessity of substantial engagement on the experimenter's part in making observations stated in terms of incompatible variables. Thermodynamic operations start traditional thermodynamic processes. The experimentalist can often observe with minimal involvement in other fields, such as astronomy.

Example

• It determine if a specific observational study provides adequate information to provide reliable estimates of causal effects.
• It develop an algebraic formulation for an estimand of a causal effect.
• It chooses parameters that would eliminate the need for randomization in trials.
• It chose a series of indirect (randomized) trials to stand in for costly or impractical direct studies.
• Poor compliance in randomized trials is used to predict (or bound) therapeutic efficacy.

The study begins with the nonparametric specification of structural equations, then offers the three principles of inference, and then proposes an operational definition of structural equations to construct the semantics required for a theory of interventions.

Results

The theory's categorization of confounding factors, or those that, if controlled for, would restore the expected causal connection between the variables of interest, has important practical implications. Any set of non-descendants of X that d-separates X from Y after deleting all arrows originating from X is demonstrated to be sufficient for calculating the causal effect of X on Y. The "backdoor" criteria give a formal description of "confounding" in mathematics and direct researchers towards measurable sets that are easily accessible.

Conclusion

Multiple researchers have proved that this calculus characterizes all post-intervention probabilities expressible in fundamental conditional probabilities. While the do-calculus necessitates some understanding of the causal structure, scholars have come up with lesser versions that don't require this. Furthermore, the causal structure that may be confidently deduced from the original probability is all that these calculi rely on.

Image source: Unsplash