What Is Denying the Consequent? The Argument Form Behind Testing Ideas

Two argument patterns involving a conditional statement are valid. Affirming the antecedent takes the conditional and the truth of its first part and concludes the second, so from the claim that rain implies wet ground and the fact of rain it follows that the ground is wet. Denying the consequent takes the conditional and the falsity of its second part and concludes that the first part is false, so from the same conditional and dry ground it follows that it did not rain. Both are valid, meaning the conclusion cannot be false while the premises are true, and both are used constantly. The second is the more powerful in practice, since it permits eliminating a claim rather than confirming one, and elimination is what advances an argument.

Two superficially similar patterns are not valid and the difference matters:

• •Affirming the consequent, concluding that it rained because the ground is wet, which ignores that something else could have wet it
• •Denying the antecedent, concluding the ground is dry because it did not rain, which makes the same mistake in reverse
• •Both errors treat a conditional as though it ran in both directions, which it does not
• •Both are extremely common in ordinary argument and in inference from evidence
• •Both become valid only if the conditional is strengthened into a biconditional, meaning each implies the other
• •Checking whether the conditional really runs both ways is the practical test for whether an argument commits them

The valid form of denial is the logical skeleton of empirical testing. A hypothesis implies an observable consequence, the observation is made, and if the consequence does not occur the hypothesis is false. That is why prediction matters more than explanation in assessing an idea, since a prediction exposes the claim to exactly this form while an explanation of something already known does not. It is also why confirmation is weaker than refutation, since finding the predicted consequence affirms the consequent and does not establish the hypothesis, which is the logical basis for the observation that theories can be refuted and never conclusively proved. Popper built an account of scientific method around precisely this asymmetry, treating falsifiability as the mark distinguishing scientific claims from others.

Everything here depends on what a conditional statement means, which is less straightforward than it appears. In formal logic a conditional is treated as false only when the first part is true and the second false, which makes it true whenever the first part is false regardless of any connection, so a statement that if the moon is cheese then arithmetic works comes out true. That is the material conditional, and it is adopted because it makes the logic work smoothly rather than because it matches ordinary usage. Ordinary conditionals seem to assert a connection, and they include counterfactuals about what would have happened, which the material reading handles badly. Several alternative accounts exist and none is settled. The valid argument forms survive all of this, since they hold on any reasonable reading, and the underlying question about what a conditional claims does not.

The clean picture is complicated by a problem that is well understood and rarely mentioned alongside it. A hypothesis never implies an observation on its own, since deriving a prediction requires auxiliary assumptions about instruments, about background conditions and about other theories, so when the prediction fails the valid conclusion is that something in the whole set is false rather than that the hypothesis specifically is. That is the Duhem-Quine problem, and it means a failed test can always be met by adjusting an auxiliary assumption instead, which is sometimes legitimate and sometimes evasion, and no rule distinguishes the two cases reliably. Historical episodes include predictions failing and being resolved by discovering an unknown planet, which vindicated the adjustment, and comparable adjustments that merely postponed the abandonment of a theory.

From a conditional and the falsity of its consequence, the antecedent is false, which is valid and is what every test does. Concluding the antecedent from the consequence is the invalid cousin and is extremely common. The asymmetry between refuting and confirming follows directly. A failed prediction shows something in the whole set of assumptions is false, not necessarily the hypothesis.