Core Principles of Critical Inquiry

Critical inquiry is the deliberate examination of a question, the information relevant to it and the reasoning used to reach an answer. It includes checking your own explanation with the same care you bring to someone else's.

Start by putting the claim into one sentence. “This is better” leaves almost everything open. “This scheduling system shortened average delivery time for this company during its first three months” identifies an outcome, a setting and a period. You can now ask what was measured and what comparison supports the claim.

Also identify the kind of claim being made. Different questions require different kinds of support.

These statements are connected, but establishing the first does not automatically establish the others. A recommendation adds judgments about what matters and what should be done.

Opening phrases do not determine truth. “I believe the workshop starts at nine” makes a checkable factual claim. “A study shows…” can introduce an accurate summary or a misleading one. A factual claim is something that can be assessed for accuracy; it is not already an established fact merely because it sounds objective.

If a statement remains vague, ask for a concrete example or an observable meaning. Agreeing on what “effective,” “fair” or “successful” means can resolve confusion before an argument develops.

In reasoning, an argument is a set of reasons offered to support a conclusion. The reasons are its premises. Evaluating an argument involves two questions: are those premises acceptable, and do they support the conclusion?^[1]

Everyday arguments often leave a connecting assumption unstated. Consider: “Most survey respondents want a quiet room, so the library should build one.” The survey result is one premise; building the room is the proposed conclusion. Several further assumptions are doing work.

An unstated assumption is not automatically a mistake. Ordinary conversation would become exhausting if every background fact had to be repeated. The useful task is to identify assumptions that are uncertain, disputed or essential to the conclusion.

Ask: “What would have to be true for this reason to support that decision?” Then make the answer explicit. In the library example, “The survey represents all users” is easier to examine than a general feeling that the proposal seems convincing.

Keep an observation separate from your interpretation. “My message has not received a reply” is an observation. “They are angry with me” is one possible explanation. Before treating the explanation as a fact, consider what connects the two and what other explanations remain plausible.

Who, what, when, where, why and how provide a practical starting checklist. They help uncover missing context, but cannot guarantee completeness or accuracy. Use the answers to identify the most consequential uncertainty and choose the next question.

The questions work together. “Who responded?” affects what a survey can tell you. “When?” helps assess a causal sequence. “Where?” limits a prediction about another setting. A polished answer to one question cannot compensate for an unanswered question that the conclusion depends on.

You do not need to investigate all six equally. For a historical photograph, origin and date may be decisive. For a claim that a new routine improves concentration, measurement and comparison may deserve most of your attention.

Some arguments aim to establish a conclusion necessarily; others offer evidence that makes a conclusion more likely. Recognizing the difference helps you ask for the right kind of support.

A deductive argument is valid when its premises cannot all be true while its conclusion is false. It is sound when it is valid and its premises are true. An inductive argument offers degrees of support rather than that guarantee.^[2]

A useful piece of evidence bears on the particular claim you are assessing. A receipt can document a purchase. An interview can explain someone's experience. A representative survey can estimate how widespread a reported experience is. These sources answer different questions.

Personal accounts deserve accurate treatment. “This happened to me” may be valuable evidence about an experience and a reason to investigate. It does not automatically establish how often the same thing happens or which factor caused it. Equally, a population average does not erase an individual's experience.

Check the original material when possible. Does the cited report actually contain the claim? Does a quotation preserve its qualifications? Do ten articles provide ten independent investigations, or are they repeating one announcement?

Looking beyond a page to investigate its source is often called lateral reading. In a 2021 study, college students improved their online source evaluations after instruction that included this approach. The study used a before-and-after design, so it should not be treated as definitive evidence about every learner or setting.^[3]

Once a source is credible, a further question remains: does this particular evidence support this particular inference? A careful report about one group may be misapplied to another. An accurately reported association may be presented as proof of a cause.

For a fuller approach to source checking, uncertainty and confidence, continue with Balancing Openness and Skepticism. Here, the central task is to identify the connection between the evidence and the conclusion.

Two things can vary together without one causing the other. A third factor may affect both; the direction of influence may differ from the proposed explanation. Observing that one event preceded another is useful information, but is insufficient by itself to establish causation.^[4]

Imagine a delivery company introduces new scheduling software. Its average delivery time falls from four days to three. The change is worth examining, but the records alone do not isolate the software's contribution.

Perhaps parcel volume fell, more staff were available or a larger share of deliveries went to nearby destinations. Perhaps the previous period contained an unusual disruption. Each possibility suggests information that could help distinguish explanations.

Well-designed experiments can help separate a proposed cause from competing influences. Other careful research designs can also contribute causal evidence; the appropriate design depends on what can be studied and under which conditions. A convincing account explains why its comparison is informative and what uncertainty remains.

Keep the conclusion aligned with the available evidence: “Delivery time improved during the period when the software was introduced; its contribution is not yet clear.” This preserves the observed improvement while identifying the unresolved causal question.

The same discipline applies to personal routines. If you began a new morning practice, changed your working hours and stopped checking messages before breakfast, an improvement in concentration cannot easily be assigned to one change. A simple record may help you notice patterns, without turning a short personal experiment into a universal claim.

A number becomes meaningful when you know what was counted, out of how many, over what period and compared with what. The examples in this section are invented to show the arithmetic.

Relative change and absolute difference

Suppose defects are found in 40 of 1,000 items before a process change and 20 of 1,000 afterward. The observed defect rate fell from 4% to 2%: a 50% relative reduction and a 2 percentage point absolute reduction. Both descriptions are correct. Neither calculation, on its own, establishes why the change happened.

Relative effects need a starting rate to show their absolute size. Cochrane's guidance makes this distinction explicit when interpreting intervention effects: the same relative effect can imply different absolute effects at different baseline risks.^[5]

Finding faults and interpreting a flag

Now imagine a scanner checks 1,000 items. Independent inspection establishes that 20 are faulty and 980 are sound. The scanner flags 16 faulty items and also flags 49 sound ones.

The scanner catches 80% of the faulty items: 16 out of 20. But only about 25% of flagged items are faulty: 16 out of 65. Those percentages use different denominators and answer different questions. “It detects 80% of faults” does not mean “a flagged item has an 80% chance of being faulty.”

In this example, a flag is a reason for further inspection. Whether the scanner is useful also depends on the consequences of missed faults, unnecessary inspections and available alternatives.

Precision and statistical language

Ask how participants or items were selected before being impressed by a large sample. Thousands of responses from a narrowly selected group may still poorly represent the population you care about. Also check whether “average” means a mean or a median, and whether important variation is hidden.

A small p-value does not measure an effect's size or practical importance, and it is not the probability that a hypothesis is true. The American Statistical Association recommends interpreting such results alongside study design, context and fuller reporting.^[6] Ask what changed, by how much, and with what uncertainty.

Fallacy names can help you recognize recurring problems, but a label needs an explanation. Say which claim was distorted, which fact is irrelevant or which connection is unsupported. Introductory accounts describe patterns such as straw man arguments and irrelevant personal attacks.^[7] Their application requires attention to the actual exchange.

Straw man: changing the position under discussion

A straw man replaces someone's position with an easier target. Suppose a colleague proposes two meeting-free hours each week. “You want people to stop communicating” exaggerates that proposal and leaves the actual question unanswered.

A fair criticism could ask how urgent coordination would work during those hours. It engages with the proposal and tests a relevant consequence.

Repair question: “Can we state the proposal in terms its supporter would recognize before evaluating it?” Check the original wording; you do not need to strengthen it into a different proposal.

Ad hominem: using an irrelevant personal criticism

An ad hominem error occurs when an irrelevant fact about a person is used to dismiss their argument. “Her booking calculation is wrong because she dresses badly” offers no reason to reject the arithmetic.

Personal information can, however, matter to the reliability of testimony. Relevant competence, a documented history of fabrication or a conflict of interest may deserve examination. Such information does not automatically refute independently supported reasoning.^[8]

Likewise, pointing out hypocrisy may be relevant to someone's conduct, but does not settle the rule they propose. A person who arrives late can still correctly observe that lateness disrupted a meeting.

Repair question: “How does this fact about the speaker affect the evidence or reasoning we are assessing?”

Misused authority: giving expertise the wrong job

Expert testimony can provide reasonable support. Its value depends on relevant expertise, accurate reporting, the evidential basis and how the claim relates to other qualified views. Credentials are not a guarantee, and expertise in one field does not automatically transfer to another.^[8]

A famous musician's preference for a concert hall tells you about that person's experience. It does not, by itself, establish a technical claim about the building's structural capacity. A qualified engineer's assessment of that capacity would be relevant evidence.

Repair question: “What specific expertise supports this claim, and what assessment or evidence is the expert relying on?” You can ask this without pretending you can personally redo every specialist investigation.

These patterns are useful prompts for closer examination. Naming one should lead to a concrete question about the argument.

Context matters. Two options can be exhaustive when the question is tightly defined. A forecast of escalating consequences can be reasonable when the causal links are supported and uncertainty is acknowledged. A slippery-slope label alone does not settle the forecast.^[8]

The following scenarios are hypothetical. They show how clearer questions can lead to a narrower, better-supported conclusion and a practical next step.

People can disagree about facts, definitions, causal explanations, predictions or priorities. Those disagreements call for different responses. More statistics will not necessarily settle a difference about how a shared space should be used.

In the library example, two people might agree that a quiet room would be popular while disagreeing about whether it deserves space currently used for group activities. Evidence helps describe likely consequences; the decision also involves whose needs to prioritize and which trade-offs to accept.

Try summarizing the other person's reasoning before responding: “You agree that noise is a problem, but think a permanent room would displace too many group activities. Have I understood?” This gives them a chance to correct the account and helps avoid arguing against an invented position.

Apply comparable standards to your own preferred option. If you question the representativeness of a survey you dislike, examine a favorable survey's recruitment too. If you demand a clear outcome measure from another proposal, define one for yours.

Ask what evidence would count against your interpretation. The answer need not be a single decisive test. It could be a pattern of observations, a corrected record or a plausible alternative explanation with better support.

Choose one claim that matters to a current decision. A few written lines can reveal more than repeatedly rereading a persuasive explanation.

Match the effort to the consequences. A small, reversible preference may need little investigation. A costly or difficult-to-reverse decision deserves more attention to the assumptions it depends on. Stop collecting material when another round is unlikely to resolve the relevant uncertainty or change the available action.

Practice also includes noticing your own explanations: “I always fail at this,” “Everyone expects me to agree,” or “This is the only way I can manage.” Ask what the words “always,” “everyone” and “only” cover, and whether a more precise description creates room for another response.

Self-Reflection Tools offers a related path for examining personal interpretations. Detecting Manipulation and Propaganda applies careful evaluation to persuasive messages and pressures on choice.

These sources support the reasoning distinctions and research statements above. Numerical examples and everyday scenarios are hypothetical teaching examples.