Definition
An argument is valid when it is impossible for its premises to be true and its conclusion false at the same time. Validity is the central virtue logic looks for: it is the guarantee that the inferential link between premises and conclusion is truth-preserving — not merely likely to preserve truth, but necessarily so in every situation where the premises hold.
Crucially, validity says nothing about whether the premises are actually true. The argument "All cats are reptiles; Felix is a cat; therefore Felix is a reptile" is valid — if its premises held, the conclusion would have to follow — even though a premise is false. Validity concerns the strength of the connection, not the truth of the starting points. Separating these two evaluations — the connection and the inputs — is the foundational move that makes formal logic possible.
Why it matters
How it works
The counter-example test
The most practical way to test an argument for validity is to try to construct a counter-example: a possible situation in which every premise is true yet the conclusion fails. If no such situation can be constructed, the argument is valid; if one can be found, it is invalid. This thought-experiment turns an abstract notion into a concrete check. It matters that the situation need only be possible, not actual — validity is evaluated across all logically conceivable states of affairs, not just the real world.
Because validity hinges on what is possible rather than what is actual, it is tied directly to logical form. Two arguments with the same form — the same arrangement of subject, predicate, and connective — stand or fall together. This is why logicians study abstract patterns rather than particular claims: securing the form secures every argument that instantiates it, across any domain of inquiry.
Deductive versus inductive validity
Not all good reasoning is equally airtight. The deductively valid inference offers an absolute guarantee: given true premises, the conclusion cannot fail. The inductively valid inference offers strong but not conclusive support: "storm clouds are gathering, so rain is likely" is a good bet, but a shift in the wind could spoil it. We reason inductively constantly — diagnosing faults, solving crimes, forming scientific hypotheses — and inductive strength is a real virtue. But formal logic has historically concentrated on the deductive case, partly because deductive validity admits of a precise definition and a mechanical test, while inductive strength is a matter of degree.
The distinction matters in practice. A scientist who says "my data supports the hypothesis" is making an inductive claim; the support is real but defeasible. A mathematician who says "the theorem follows from these axioms" is making a deductive claim; if the argument is valid and the axioms hold, the conclusion is beyond dispute.
How categorical logic makes validity mechanical
Aristotle's syllogistic — the formal system introduced in a complete introduction to logic — turned validity testing from a judgement call into an algorithm for the first time. Categorical propositions take four standard forms: all S are P, no S are P, some S are P, some S are not P. From these, a syllogism is an argument with two premises and one conclusion, each a categorical proposition, sharing terms in a specific pattern. The Venn diagram test maps the premises as shadings and markings on three overlapping circles representing the three terms; if the conclusion is already visible in the diagram after the premises have been marked, the argument is valid. Five formal rules offer an equivalent mechanical check.
What this system makes vivid is the independence of form and content. An argument about engineers, meetings, and punctuality has the same logical structure as an argument about mammals, fish, and gills. Once the form is secured as valid, the conclusion follows from the premises regardless of what the terms refer to. Validity has stopped being intuition and become computation.
Truth functions and the limits of the guarantee
Propositional logic — where whole sentences are connected by and, or, and not — extends validity testing to a larger class of arguments via truth tables. An inference is valid if no row in its truth table makes every premise true and the conclusion false. This is a complete test: if no such row exists, the argument is valid without exception.
But the truth-table method also reveals a class of vacuously valid arguments — arguments whose premises are jointly impossible. If the premises can never both be true, there is no row to fail, so the argument is trivially valid. The troubling inference "p and not-p, therefore the moon is made of cheese" passes the truth-table test. This is not a defect in the definition; it follows directly from what validity means. But it is a reminder that validity is a property of the inferential link, not of the content, and that a vacuously valid argument is useless as evidence for its conclusion.
Validity in a richer space of truth values
Classical logic assumes that every statement is either true or false. Self-referential statements — "this sentence is false," Russell's set of all sets that are not members of themselves — seem to be neither, or both. One response is to allow four truth-value possibilities: true only, false only, both, or neither. Validity keeps its definition: no situation makes the premises true and the conclusion not true. But with richer situations available, some inferences that seemed valid in the classical case come out invalid. The troubling chain "q, not-q, therefore p" — from any contradiction, any conclusion follows — is now invalid: a situation where q is both true and false but p is only false makes both premises true while leaving the conclusion not true. This is a deeper diagnosis of why the inference has always felt wrong.
Internal and external validity in research (Psychology)
In empirical research, validity bifurcates into two distinct concerns that can trade off against each other. Internal validity is about the inside of a study: does the design support the causal claim the researchers are making? A randomly assigned experiment where one variable is manipulated and all others are controlled has high internal validity. If workers were simply observed in offices that already had plants versus offices that did not, there is no internal validity for a causal claim — the managers who chose to install plants may also pay better, manage differently, and hire differently. External validity is about whether the finding travels: would the same result appear in different populations, settings, and time periods? A tightly controlled laboratory study may have high internal validity but low external validity if the conditions are too artificial to generalise.
The two pull in opposite directions. Strict experimental control increases internal validity but often creates artificial conditions that reduce external validity. Field studies preserve ecological realism but surrender the control needed for causal inference. Good research design is the management of that tension rather than the elimination of it.