Critical reasoning: arguments and inferences

In short

Critical reasoning tests whether you can take a short passage, separate what is *claimed* (conclusion) from what *supports* it (premises), and judge what follows, what is taken for granted, and what would make the claim stronger or weaker. Unlike syllogisms it uses everyday prose, not formal "All A are B" statements, so the trap is real-world knowledge rather than logic alone. It is examined in CAT and XAT verbal ability and logical reasoning, in SNAP, NMAT, CMAT, GRE and GMAT, in bank PO and SSC English/reasoning papers, and in the verbal section of placement tests from TCS NQT, Infosys, Wipro, Capgemini, Accenture, Cognizant and AMCAT/eLitmus. In interviews and on the job the same skill appears as case questions, guesstimates and reviewing a business claim in a report.

Explained

What an argument is, and what it is not

• An argument is a set of statements where one or more premises are offered as reasons to accept a conclusion. Without that support relationship it is not an argument, only a set of assertions or a description.
• Premise: evidence, data, observation, or an accepted claim. Conclusion: the point the author wants you to accept.
• Premise indicators: because, since, as, for, given that, in view of, due to, the reason is, studies show, evidence suggests.
• Conclusion indicators: therefore, hence, thus, so, consequently, it follows that, clearly, this shows that, we must, should.
• When no indicator words exist, use the "Therefore test": insert "therefore" before each sentence; the one that reads naturally as the pay-off is the conclusion.
• The main conclusion is not always the last sentence. An intermediate conclusion (sub-conclusion) is supported by premises and in turn supports the main conclusion.
• Counter-premise: a statement the author concedes but overrides - usually introduced by although, however, but, while it is true that, despite, yet. The main conclusion almost always follows "but/however".
• A fact can be verified; an opinion/claim cannot. Options stating opinions are rarely valid inferences.

Deduction, induction and the structure of support

• Deductive argument: if the premises are true the conclusion *must* be true (valid). Validity is about form, truth is about content - a valid argument can have false premises. Sound = valid + true premises.
• Inductive argument: premises make the conclusion *probable*, not certain - generalisation from a sample, analogy, causal reasoning, statistical inference. Judged as strong or weak, not valid or invalid.
• Abductive reasoning (best explanation): given an observation, choose the hypothesis that explains it best; the staple of "which option explains the discrepancy" questions.
• Analogy-based argument: X resembles Y in some respects, so what is true of Y holds for X. Attacked by showing a relevant dissimilarity.
• Causal argument: A is followed by B, so A causes B. Attacked by (i) reverse causation, (ii) a third common cause, (iii) coincidence, (iv) the sample being unrepresentative.
• Sample-based argument: strong only if the sample is large enough, randomly drawn and representative of the population the conclusion is about.

Inference: what "must be true" means

• An inference is a statement that must be true, or is very strongly supported, *given only the passage*. It is not a summary, not the conclusion restated, and not something the author merely implies rhetorically.
• Test every option against: Is it 100% backed by the text? If you must add any outside fact, it is not an inference.
• Safe inferences are small, weakly worded steps: "at least some", "may", "need not", "not all". Unsafe options carry extreme words: all, only, never, always, must, impossible, best, the sole cause.
• Scope shift is the commonest trap: the passage speaks of one city, one year or one firm; the option speaks of the country, every year or the industry.
• Direction and comparison traps: the passage gives absolute numbers, the option gives percentages (or the reverse); or the passage compares A with B and the option compares A with C.
• Negative inference: if the passage gives a complete list of possibilities and rules out all but one, the remaining one is inferable.
• For conclusion/main-idea questions the answer summarises the whole passage; for inference questions the answer is a *new* statement derived from it. Mixing the two costs marks.

Assumptions, and the question types built on them

• An assumption is an unstated premise the argument needs. A necessary assumption is required for the argument to work at all; a sufficient assumption would make the conclusion follow completely. Placement papers almost always mean *necessary*.
• Negation (denial) test: negate the candidate assumption. If the argument collapses, it is a necessary assumption; if the argument survives, it is not.
• Strengthen: adds a new fact that makes the conclusion more likely - confirms the assumption, rules out an alternative cause, shows the sample was representative.
• Weaken: attacks the link between premise and conclusion - gives an alternative explanation, shows the sample biased, provides a counter-example, shows the cause absent yet the effect present.
• Flaw / describe the error: name the reasoning defect (see fallacies) rather than disputing the facts.
• Evaluate the argument: pick the question whose *two possible answers* push the conclusion in opposite directions - that is the test of usefulness.
• Resolve the paradox / explain the discrepancy: find a fact consistent with both halves of the apparent contradiction.
• Boldface / role of a statement: label each highlighted part as premise, intermediate conclusion, main conclusion, counter-premise or background.
• Parallel reasoning: match the *structure* (pattern of premises and conclusion), ignoring the subject matter.
• Complete the passage: choose the option that follows the last connector's logic - "however" needs a contrast, "therefore" needs a consequence.

Common fallacies examined

• Post hoc ergo propter hoc: treating sequence as causation. "Sales rose after the new logo, so the logo raised sales."
• Correlation taken as causation: two variables move together; a third factor may drive both.
• Hasty generalisation: a conclusion about a population from a tiny or self-selected sample (an online poll, two friends' experience).
• Sampling / survey bias: non-response bias, leading questions, surveying only existing customers (survivorship bias).
• Circular reasoning (begging the question): the conclusion restated as a premise.
• Ad hominem: attacking the speaker instead of the claim. Appeal to authority: citing fame rather than relevant expertise. Appeal to popularity (bandwagon): many believe it, so it is true.
• Straw man: refuting a distorted version of the opponent's claim.
• False dilemma (either/or): presenting two options when more exist.
• Slippery slope: assuming a first step inevitably leads to an extreme outcome without showing the chain.
• Equivocation: using one word in two senses (e.g. "free" as costless and as unrestricted).
• Composition / division: what is true of parts must be true of the whole, or the reverse - "every player is excellent, so the team is the best".
• Percentage vs base fallacy: a large percentage rise on a tiny base, or ignoring that the population itself grew.
• Confusing necessary and sufficient conditions: "you need a degree to apply" does not mean a degree gets you the job.
• Appeal to ignorance: no evidence against X, therefore X is true.

A reliable solving method

1. Read the question stem first - it tells you whether to hunt for a conclusion, an assumption, a weakener or an explanation.
2. Read the passage once and mark premises and conclusion; paraphrase the argument in your own words in under ten words.
3. State the gap between premise and conclusion before looking at the options - this pre-empts the trap options.
4. Use elimination: strike options that are out of scope, too extreme, irrelevant, already stated in the passage, or that answer a different question type.
5. For strengthen/weaken, apply the relevance test - if the option were true, does the conclusion become more or less likely? No change means wrong.
6. Accept the passage as true even if you disagree; never import real-world knowledge, company experience or current affairs.
7. Watch the qualifiers in options - one word (all, only, never, mainly) usually decides right from wrong.
8. Time discipline: in CAT-style sets roughly 90-120 seconds per question; in TCS/AMCAT verbal sections 45-60 seconds. Attempt single-question passages before long sets.

Argument-based question formats in Indian papers

• CAT and XAT: para-summary, inference, conclusion from a paragraph, assumption, and XAT's dedicated critical-reasoning plus a decision-making section with ethical dilemmas.
• Bank PO (IBPS/SBI) mains: statement-and-assumption, statement-and-conclusion, cause and effect, course of action, and "critical reasoning" sets of strengthen/weaken arguments.
• Placement aptitude (TCS NQT, Infosys, Wipro Elite, Accenture, Capgemini, AMCAT, eLitmus): short passages with conclusion or inference options, often labelled "definitely true / probably true / data inadequate / probably false / definitely false".
• Five-way inference grading used in many bank and PSU papers: definitely true (follows from the passage), probably true (likely but not certain), data inadequate, probably false, definitely false (contradicts the passage).
• UPSC CSAT Paper II and state PSC papers include comprehension-based assumption and inference questions; GATE has a General Aptitude section (15 marks) whose verbal and analytical aptitude items include such questions.

Key points

• Argument - premises offered in support of a conclusion; no support relationship means no argument.
• Conclusion - the claim being argued for; find it with the "therefore test", not by position in the passage.
• Counter-premise - a concession before "but/however"; the main conclusion usually comes after it.
• Validity - form guarantees the conclusion from the premises; soundness - valid plus true premises.
• Induction - probable, not certain; judged strong or weak, and weakened by unrepresentative samples.
• Inference - must be true from the text alone; prefer weak, cautious wording over sweeping statements.
• Scope shift - the commonest wrong option: changes place, time, group or degree from the passage.
• Necessary assumption - the argument fails without it; confirm with the negation test.
• Sufficient assumption - added to the premises, it would make the conclusion follow; rarely the required answer.
• Strengthen - rules out alternative explanations or confirms the assumption; weaken - supplies an alternative cause or counter-example.
• Post hoc fallacy - sequence mistaken for causation; the most frequently tested flaw.
• Correlation vs causation - co-movement may stem from a common third factor or reverse causation.
• Hasty generalisation - population-level claim from a small or self-selected sample.
• False dilemma, straw man, ad hominem, circular reasoning - the four most commonly named fallacies in option sets.
• Percentage vs absolute - a rise in share can occur while the absolute number falls, and the reverse.
• Necessary vs sufficient condition - "required for" is not "enough for"; a frequent trap in both options and interviews.
• Resolve-the-paradox - the answer must be compatible with both conflicting facts, not deny either.
• Parallel reasoning - match structure, not topic; subject-matter similarity is a decoy.

Formulas and examples

Critical reasoning has no numeric formulas; the equivalent tools are structural tests, written here as rules.

Example 1 - Find the conclusion and the assumption. Passage: "Companies that offer work-from-home report 20% lower attrition. Our firm's attrition is high. We should therefore allow work-from-home."

• Premises: (i) WFH firms report 20% lower attrition; (ii) our attrition is high.
• Conclusion: we should allow WFH.
• Gap: the correlation in other firms is assumed to be causal and transferable to us.
• Necessary assumption: WFH is at least part of the reason for the lower attrition in those firms (and nothing about our work prevents WFH).
• Negation check: "WFH has no bearing on their lower attrition" - the recommendation loses its only support, so the assumption is necessary.
• Strengthener: firms that introduced WFH saw attrition drop afterwards. Weakener: those firms also pay 30% above market, which explains the retention.

Example 2 - Must be true (inference). Passage: "Of the 1,200 students placed from a Pune engineering college in 2025, 480 joined IT services firms and 300 joined core engineering firms. No student joined more than one firm." Which must be true?

• (a) Most students joined IT services - 480/1,200 = 40%, not a majority. Wrong (percentage trap).
• (b) At least 420 students joined firms that are neither IT services nor core engineering - 1,200 − 480 − 300 = 420. Correct, and exact.
• (c) IT services offer better pay - outside information. Wrong.
• (d) Core engineering hiring has declined - no earlier year is given. Wrong (no baseline).

Answer: (b).

Example 3 - Weaken the argument. Passage: "After our college introduced a mandatory aptitude class in 2024, the placement rate rose from 62% to 78%. The class clearly caused the improvement."

• (a) Students found the class useful - opinion, weak support.
• (b) In 2024 the number of recruiters visiting campus doubled - supplies an alternative cause for the rise. Strongest weakener.
• (c) The class was held on Saturdays - irrelevant.
• (d) Another college with no such class saw placements fall - this *strengthens* the claim.

Answer: (b) - the classic post hoc attack: find a second change in the same period.

Example 4 - Resolve the discrepancy. Passage: "A mobile brand cut its phone's price by 15% in India, yet its total revenue from that model rose by 10% in the next quarter."

• Required: a fact consistent with both the price cut and the revenue rise.
• Correct type of answer: unit sales rose by nearly 30%, so price × quantity increased (1.10 ÷ 0.85 ≈ 1.294, i.e. a ~29% volume rise lifts revenue about 10%).
• Wrong types: "the brand spent more on advertising" (explains sales effort, not the arithmetic), "customers prefer cheaper phones" (restates the premise).

Example 5 - Conditional reasoning. Passage: "If a candidate clears the coding round, she is called for the interview. Priya was not called for the interview."

• Valid inference (modus tollens): Priya did not clear the coding round.
• Invalid: "Priya was called, so she cleared the round" - she may have been called through a referral route; the statement does not say *only* coding-round clearers are called (affirming the consequent).

Where it is used

• Entrance and recruitment exams: CAT/XAT/SNAP/NMAT verbal and logical reasoning; IBPS and SBI PO mains reasoning; SSC, RRB and state PSC papers; GATE general aptitude; CLAT legal and logical reasoning.
• IT and consulting placement tests and interviews: TCS NQT, Infosys, Wipro, Accenture, Cognizant, Capgemini, Deloitte, EY, KPMG, PwC - case and guesstimate rounds are critical reasoning applied to business data.
• Analyst and research roles: business analysts, market-research and equity-research associates, policy researchers at think tanks, and data analysts who must separate correlation from causation before writing a recommendation.
• Law, journalism and compliance: drafting and attacking arguments, fact-checking, audit observations, and RTI or grievance replies where a claim must be supported by evidence on record.
• Product and operations work: reading A/B test results, root-cause analysis (five whys), post-incident reviews where "the deploy preceded the outage" must be tested, not assumed.
• Everyday professional writing: reviewing a vendor's claim, a sales deck, a news report on a survey, or an advertisement's statistics before acting on it.

Interview questions

• Q: What is the difference between a premise and a conclusion? A: The premise is the evidence or reason offered; the conclusion is the claim it is meant to support. A quick check is to insert "therefore" before a sentence - it reads naturally only before the conclusion.
• Q: How is an inference different from a conclusion? A: A conclusion is the point the author argues for in the passage; an inference is a new statement that must be true given the passage, even if the author never states it.
• Q: What is a necessary assumption and how do you confirm one? A: An unstated premise the argument needs to work. Negate it - if the argument falls apart, it is necessary; if the argument still stands, it is not.
• Q: Can a valid argument have a false conclusion? A: Yes, if a premise is false. Validity concerns only the form of the inference; soundness requires both validity and true premises.
• Q: Give an example of correlation mistaken for causation. A: Cities with more ice-cream sales see more drowning deaths; hot weather raises both, so the correlation is driven by a common third factor, not by ice cream.
• Q: What are the standard ways to weaken a causal argument? A: Offer an alternative cause, show reverse causation, show the effect occurring without the cause, or show the sample was biased or too small.
• Q: Differentiate necessary and sufficient conditions with a workplace example. A: A valid driving licence is necessary to be hired as a driver but not sufficient; clearing every round of selection is sufficient for an offer but a licence alone is not.
• Q: What is the post hoc fallacy and where have you seen it at work? A: Treating "B followed A" as "A caused B" - for instance crediting a website redesign for a sales rise that coincided with a festival-season discount.
• Q: A survey of 200 visitors to a company's website finds 90% satisfaction. What is wrong with concluding customers are satisfied? A: The sample is self-selected and covers only visitors, not all customers - self-selection and non-response bias make the 90% unrepresentative.
• Q: What is a straw man argument? A: Replacing the opponent's actual claim with an exaggerated or distorted version and refuting that instead, so the real claim is left untouched.
• Q: "If it rains, the match is cancelled. The match was cancelled. So it rained." Is this valid? A: No - it affirms the consequent. The match could have been cancelled for another reason; only "it did not rain, given the match was not cancelled" style reasoning (modus tollens) is valid.
• Q: In a strengthen question, why is restating a premise a wrong answer? A: Strengthening requires new information that makes the conclusion more likely; repeating what is already given adds no support to the premise-conclusion link.

Common mistakes

• Using personal or general knowledge instead of the passage - the passage must be accepted as true and treated as the only evidence.
• Picking the option that is factually sensible rather than the one logically supported, especially on familiar topics like education or pollution.
• Choosing options with extreme quantifiers (all, only, never, best, sole cause) for inference questions, where cautious wording is almost always right.
• Treating a restatement or summary of the passage as an inference or as a strengthener.
• Mistaking an intermediate conclusion for the main conclusion, or treating the last sentence as the conclusion by default.
• Confusing strengthen/weaken with assumption, or attacking a premise's truth when the question asks about the flaw in reasoning.
• Mixing percentages with absolute numbers, and ignoring base effects or population changes between the premise and the conclusion.
• Not reading the question stem first, then re-reading the passage repeatedly and running out of time in the verbal section.

Related topics

Statements and assumptions · Statements and conclusions · Cause and effect and course of action · Syllogisms · Data sufficiency · Reading comprehension and para summary