Logic and reasoning · An introduction
Induction vs Deduction
A valid deduction leaves no case in which its premises are true and its conclusion false. Inductive reasoning can support a conclusion about an unobserved case without ruling out every exception. Start by asking what the premises establish and how the conclusion goes beyond them.
Updated
Compare the standard of support, then check the premises
| Question | Deductive evaluation | Inductive evaluation |
|---|---|---|
| What are you testing? | Whether the conclusion follows necessarily from the premises. | How well the premises support an expectation, estimate or generalization. |
| Can true premises accompany a false conclusion? | Not in a valid deduction. A counterexample shows invalidity. | Yes. Even useful evidence can leave an unobserved exception possible. |
| What can go wrong? | An invalid inference, or an unsupported or false premise in an otherwise valid form. | A biased sample, a relevant change, missing cases or a conclusion stronger than the evidence. |
| What does true evidence add? | True premises plus validity make the argument sound. | Truth of the premises matters, but their relevance and strength still need assessment. |
| Is direction enough? | Deduction can move from general claims to general claims, too. | Induction can predict one new case rather than a universal rule. |
| What is a useful verdict? | Valid or invalid; then assess whether the premises are true. | Explain the evidence's strength, limitations and the scope of the conclusion. |
Evaluate the support supplied by the premises
Begin by identifying what the speaker offers as reasons and what the speaker wants you to accept. Those are the premises and conclusion. Then ask a conditional question: if these premises were true, what would they establish? This separates the argument's support from a speaker's confidence, the appeal of the conclusion and the actual reliability of the evidence.
For deductive evaluation, the central test is validity: there must be no interpretation in which all the premises are true and the conclusion false. For inductive evaluation, ask whether the evidence gives a good basis for the proposed conclusion despite remaining uncertainty. A confident tone, the word 'therefore' or a formal-looking diagram does not settle either test.
The distinction is commonly taught with the labels 'deductive' and 'inductive', but philosophers disagree about exactly how those labels classify ordinary arguments. Some accounts emphasize an arguer's intention; others emphasize the actual relationship between statements. Here the emphasis is on evaluating support. An unsuccessful attempt at a deduction is not automatically a good induction.
Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy · Stanford Encyclopedia of Philosophy
Three readings of one fictional file-label case
Suppose a fictional archive has the rule: every preview-only file carries label P. A particular file is preview-only. Therefore it carries P. If both premises are true, the conclusion cannot be false. This is a valid deduction. It does not independently establish that the archive follows the rule or that the particular file was classified correctly.
Now reverse the second part: every preview-only file carries P; this file carries P; therefore it is preview-only. The conclusion does not follow. A published file might also carry P. That possible file satisfies both premises and falsifies the conclusion. The problem is the reversed inference, even if the particular file happens to be preview-only.
Finally, suppose the only evidence is that twelve preview-only files inspected yesterday carried P. The proposed conclusion is that other preview-only files probably carry it too. This is evidence-based generalization rather than a guarantee. The inspection's coverage, selection, missing cases and any changes to the labeling process matter. The twelve observations do not by themselves establish the universal rule used in the first version.
Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy · Stanford Encyclopedia of Philosophy
Why general-to-specific is an unreliable shortcut
Deduction is often introduced as moving from a general rule to a particular case, and induction as moving from particular cases to a general rule. Those patterns describe some examples, but they are not a complete test. Consider: all training files are archived; all archived files have an identifier; therefore all training files have an identifier. This valid argument stays at the level of general statements.
An induction can also remain about a particular case. If a delivery has arrived before noon on most recent days, someone may predict that tomorrow's delivery will arrive before noon. The conclusion concerns one unobserved delivery, not every delivery forever. Look at the kind of support the premises provide rather than counting whether the sentences contain words such as 'all' or a person's name.
Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy
A single investigation can use both kinds of reasoning
In an invented record, eighteen of twenty deliveries arrived before noon. From the stated count, eighteen divided by twenty gives ninety percent of those recorded deliveries. The arithmetic describes the record. It does not establish a ninety-percent chance for tomorrow's delivery. Moving to tomorrow requires assumptions about whether the observations are relevant to tomorrow's route, schedule and conditions.
Now add a stipulated arrangement: if tomorrow's delivery arrives before noon, someone will be available to sign for it. If you were given that it does arrive before noon, you could deduce that someone will be available. But an uncertain prediction about arrival does not become certain simply because the next step has a valid conditional form. Keep the status of the input visible at every step.
This is why arguments should be divided into stages. A calculation about an observed sample, a prediction about a new case and a conditional consequence of that prediction can belong to the same task. Mark each conclusion and ask which premises support it. Calling the whole investigation 'deductive' or 'inductive' can hide the weakest link.
Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy
A true conclusion can still have weak support
Suppose the reverse file-label argument happens to reach the truth: the labeled file really is preview-only. That fact does not remove the counterexample to its form. An invalid route can reach a true answer by accident or through information that was never stated in the premises. To repair the reasoning, add a justified premise or use a different inference; do not treat the fortunate outcome as proof of validity.
An inductive prediction can fail without showing that every use of induction is worthless. Some uncertainty is part of the claim. The useful questions concern whether the evidence was relevant, whether exceptions were acknowledged, and whether the conclusion was appropriately limited. Equally, adding 'probably' to a poorly supported conclusion does not manufacture evidence.
Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy · Stanford Encyclopedia of Philosophy
A practical sequence for evaluating an argument
Write the premises separately from the conclusion. Make the conclusion's scope explicit: this case, the observed sample, the next case, most cases or all cases. Check whether a possible true-premise, false-conclusion case defeats a claimed guarantee. When the evidence is offered as support for an expectation, ask what makes the observed cases relevant to the unobserved ones.
Then examine the premises themselves. A valid form cannot certify an inaccurate record. A large sample does not necessarily repair a selection problem. Explain your verdict in a sentence that identifies the actual gap: 'the label could have another use', 'the record covers only one route', or 'the forecast depends on conditions staying relevant'. That explanation is more useful than attaching a method's name alone.
Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy
Check the reasoning
Can a deduction have a true conclusion but still be invalid?
Yes. The reverse file-label argument may happen to identify a preview-only file correctly, while its premises still permit a published file carrying P. Validity asks whether the premises guarantee the conclusion across the relevant interpretations, not whether this one outcome is correct. A true conclusion and a valid argument are different achievements.
Do twelve matching observations prove a rule for every file?
Not by themselves. They establish what was observed, assuming the record is accurate. A broader claim needs a basis for treating the inspected files as relevant to the uninspected ones. Selection, missing cases and changed labeling procedures can matter. A deduction from an independently justified universal rule would be a different argument.
Does 'probably' make an argument inductively strong?
No. It marks uncertainty in the wording, but does not supply relevant evidence. Explain why the stated observations support this conclusion and how far that support extends. A vague possibility, a guess and a well-supported prediction are not interchangeable merely because each avoids the word 'certain'.
Does an eighteen-out-of-twenty record mean tomorrow has a ninety-percent chance?
The division gives ninety percent of the twenty recorded cases. A probability for tomorrow requires a model and assumptions connecting that record to tomorrow. The arithmetic alone does not establish those assumptions. Keep an observed frequency, a model-based probability and a particular future outcome separate.
Must every ordinary argument fit one uncontested category?
No single categorical account is assumed here. The IEP article discusses disagreements about intention, necessity, validity and formalization as classification tests. You can still evaluate whether the premises establish the conclusion, provide qualified support, or do neither. State the support and its limits rather than treating a label as the final verdict.
Sources and further reading
References checked on 2 November 2026. Examples are constructed for explanation.
- Deductive and Inductive Arguments
Timothy Shanahan, Internet Encyclopedia of Philosophy
Selected introduction and sections on necessity, validity, formalization, general-to-particular definitions and evaluation. The article also challenges simple categorical classifications; its proposals are not all presented here as a consensus.
- Classical Logic
Stanford Encyclopedia of Philosophy
Introduction and the complete semantics section: formal language, derivability, interpretation, validity and satisfiability. The explanations below use small original examples rather than reproduce the entry's proofs.