Evidence, Generalization & Counterexamples
A single counterexample destroys a universal claim — but it does not prove the opposite. Getting that asymmetry right is one of the most underrated moves in clear thinking.
🎯 By the end you'll understand
How to evaluate empirical premises — checking evidence quality, sample size and representativeness, claim type (universal vs. probabilistic), counterexamples, anecdotal evidence, and alternative explanations. Plus the crucial asymmetry: why one counterexample kills a universal claim but doesn't prove its negation.
FOUNDATIONWhat makes a premise empirical?
Arguments contain two kinds of premises. Logical or definitional premises are true by definition or by the rules of a formal system ("All bachelors are unmarried"). They can be checked by analysing meanings and following rules. Empirical premises make claims about the actual world — claims that could, in principle, be verified or falsified by observation and evidence.
The premise "Whenever someone profits from investment, another person loses" is empirical. It is not true by definition. It is a factual claim about how financial transactions work in the real world — and it is a claim that could be tested, confirmed, or refuted by looking at actual investment outcomes.
Evaluating an empirical premise asks a different set of questions than evaluating a logical inference. The questions are:
- What evidence supports this claim?
- How was that evidence collected — is the sample large enough, and is it representative?
- Is the claim universal ("every X") or probabilistic ("X tends to")?
- Does a counterexample exist? If so, what exactly does it defeat?
- Is the evidence anecdotal or systematic?
- Are there alternative explanations for the evidence that has been offered?
We'll work through each of these with our running example.
P1: "It is wrong to profit when another person loses."
P2: "Whenever someone profits from investment, another person loses."
C: "Therefore, investing is wrong."
P1 is a moral premise — a value claim we'll set aside here. P2 is our empirical target for this lesson.
CLAIM TYPEUniversal vs. probabilistic claims
Before you look for evidence, identify what kind of claim you're actually being asked to evaluate. This step is more important than it looks, because the logical standard for evidence is different in each case.
| Claim type | Form | Evidence standard | Defeated by |
|---|---|---|---|
| Universal | "Every X is Y" / "Whenever X, Y" | Must hold for all cases | A single counterexample |
| Probabilistic / statistical | "X tends to Y" / "Most X are Y" / "X correlates with Y" | Must hold for a significant majority or proportion | Requires systematic contrary evidence, not one exception |
| Existential | "Some X is Y" / "At least one X is Y" | Must hold for at least one case | Impossible to defeat by counterexample — requires showing no case exists |
P2 in our running example uses the universal form: "Whenever someone profits from investment, another person loses." The word "whenever" is a universal quantifier — it is not saying "often" or "usually." It is saying every single time.
This matters enormously. A universal claim is logically fragile — one exception breaks it entirely. A probabilistic claim is much more durable — one exception is just a data point in a distribution.
When you encounter a strong empirical premise, isolate the quantifier word. Every, all, always, whenever, never, no — these are universal. Most, many, tends to, often, generally, correlates — these are probabilistic. The quantifier tells you the evidence standard the claim must meet.
CORE TOOLThe counterexample
A counterexample is a specific case that satisfies the "if" part of a universal claim but falsifies the "then" part. It works against universal claims precisely because those claims have no room for exceptions.
Consider P2: "Whenever someone profits from investment, another person loses."
Here is a counterexample: A company raises capital by issuing new shares. The founders invest in productive machinery. The business grows, creating new goods and services. Early investors profit — not because any other investor lost, but because new value was created. The economy's total wealth grew. This is a case where someone profited from investment and no one experienced an equivalent loss. P2 is false as a universal claim.
It satisfies "someone profits from investment" (the condition) and clearly fails "another person loses" (the claimed consequence). The total wealth in the scenario grew — new machinery, new output, new wages. There is no matching loss to locate anywhere in the transaction.
A depositor earns interest from a bank. The bank lent that money to a business that expanded and generated profit greater than the interest owed. The depositor gains; the business owner gains; the bank earns its spread. Total wealth grew. Three parties gained; none lost a matching amount.
KEY INSIGHTThe crucial asymmetry: what a counterexample does and doesn't prove
Here is the most commonly mishandled step. When you find a counterexample to a universal claim, you have established exactly one thing: the universal form is false. You have not established anything beyond that.
Finding a case where an investor gained without anyone else losing proves that P2 — "Whenever someone profits from investment, another person loses" — is false as a universal claim. That's it. That's the full scope of the counterexample's logical work.
It does not prove that "investment never harms anyone." That would be the opposite overgeneralization — equally universal, equally likely to be false. Some investment gains do come at others' expense: a successful short seller profits precisely because the stock price fell; a landlord in a gentrifying area profits while long-term residents are priced out. These are real cases.
The counterexample narrows the claim to something that might still be worth debating. Instead of the original universal "Whenever someone profits, someone loses," we might explore:
- "Investment profits sometimes come at others' expense" — a weak existential claim. Almost certainly true and probably uncontested.
- "Investment profits often come at others' expense" — a strong probabilistic claim. Now we need systematic evidence about the proportion of cases, not just one example.
- "In financial markets with fixed-supply assets (like existing stock shares or existing housing), gains are more zero-sum than in productive investment" — a domain-specific claim. Probably closer to true and genuinely interesting.
The counterexample moved us from an overreaching universal to a more precise, more defensible set of claims. That is exactly what it is supposed to do.
Once P2 is narrowed — say, to "Investment sometimes harms others" — the original argument's conclusion, "Therefore, investing is wrong," no longer follows. You would need a much stronger moral and empirical theory to get from "sometimes causes harm" to "always wrong." The whole argument structure has to be rebuilt from the corrected premises up.
EVIDENCE QUALITYSample size, representativeness, and anecdote
Even when a claim is probabilistic — hedged with "often" or "tends to" — the evidence offered still needs to be evaluated for quality. A universal claim is easier to test (one counterexample suffices); a probabilistic claim requires a representative body of evidence.
Sample size
A sample is the set of cases from which a generalization is drawn. Small samples allow large statistical fluctuations — you can easily draw ten black balls from an urn that is 60% white, just by chance. The smaller the sample, the wider the uncertainty around any proportion it appears to support.
Representativeness
Even a large sample misleads if it was not drawn from the right population. Convenience samples — surveying only people who respond to an online poll, studying only hospitalized patients, measuring only liquid markets — produce biases that systematic sampling tries to correct. Before accepting a statistic, ask: who or what is in this sample, and who or what is missing?
Anecdotal evidence
An anecdote is a single case, usually vivid and memorable. Anecdotes are not worthless — they can establish that a phenomenon exists (existential claim), generate hypotheses, or illustrate a mechanism. What they cannot do is support a universal or high-probability claim. "My uncle invested in stocks and lost everything" is evidence that losses happen; it is not evidence that losses are typical.
The availability heuristic (lesson 2.1) makes vivid individual stories feel more representative than they are. A memorable case of investment-caused harm looms larger in judgment than drier base-rate data. This is a known cognitive failure mode — not a feature.
CRITICAL STEPAlternative explanations
Even strong, representative evidence that X and Y are associated does not by itself establish that X caused Y — or that the only mechanism is the one claimed. Before accepting the inference behind a premise, ask whether an alternative explanation accounts for the data equally well.
In the context of P2 ("investment profits require someone else's loss"), the evidence that is real — that some investment gains are indeed at others' expense — can be explained in multiple ways:
- Zero-sum financial instruments: Derivatives, short selling, and certain market structures are designed so one party's gain mirrors another's loss. This is genuine. But it applies to a subset of investment activity, not to all productive capital allocation.
- Wealth concentration without destruction: If rich investors earn higher returns than small investors, wealth inequality grows — but this does not require that small investors lost an amount equal to what rich investors gained. A rising tide that lifts all boats, but lifts some more than others, is not zero-sum.
- Externalities: Some investments create harms for third parties — environmental damage, displacement, labour exploitation — that are real costs even if they don't appear on a ledger. These are genuine ethical concerns, but they are an argument about externalities, not about the zero-sum structure of the transaction itself.
Distinguishing these explanations matters because each one has different policy implications and different moral weight. Conflating them — treating all investment as zero-sum because some investment harms others — is an error in reasoning, not just an error in values.
APPLIEDBack to our argument: a full evidence audit of P2
Let's apply every tool in this lesson to P2: "Whenever someone profits from investment, another person loses."
| Question | Answer for P2 |
|---|---|
| What evidence supports it? | Real cases exist where gains were zero-sum (short sales, fixed-pie negotiations, some market trades). But "some cases" does not support "whenever." |
| Sample quality? | Typically supported by vivid anecdotes or by cherry-picking zero-sum instruments. Not a representative sample of all investment activity. |
| Universal or probabilistic? | Universal ("whenever") — logically fragile. One genuine counterexample suffices to defeat it. |
| Counterexample available? | Yes: productive investment in new capacity creates new value. Counterexample defeats the universal form. |
| What does the counterexample prove? | Only that the universal is false. Not that investment never harms anyone. |
| Alternative explanations? | Where harms do occur: externalities, market structure, wealth-inequality dynamics — not necessarily zero-sum by definition. |
| Verdict | P2 is false as stated. It should be replaced with a narrower, evidence-based claim before the argument can be taken seriously. |
With P2 defeated, the conclusion "Therefore, investing is wrong" does not follow from the remaining premises. The argument's logical structure is intact — the inference is valid — but one of the foundations has crumbled. To rebuild the argument honestly, you would need a premise that accurately describes when and how investment causes harm, and a moral theory connecting that harm to a verdict on investing.
INTERACTIVEClaim stress-tester
For each strong claim below, work through three steps: the strongest supporting example, a genuine counterexample, and the narrower version of the claim that survives the counterexample. Then reveal the model answer.
PRACTICE🔨 Weekly exercise
For every strong universal or near-universal claim you encounter this week — in conversation, news, social media, or an argument someone makes — construct three things:
- The strongest supporting example: What real case best supports the claim as stated?
- A genuine counterexample: What real case falsifies the universal form?
- The narrower surviving version: What more precise claim does survive the counterexample?
You don't need to do this out loud or confrontationally — do it in a notebook or mentally. The goal is building the habit of asking "is this claim actually universal, and if so, does one exception exist?" before accepting or rejecting it wholesale.
WEAKNESSCommon mistakes to avoid
"Investment doesn't always cause harm" — established by counterexample. "Investment never causes harm" — a completely separate universal claim requiring its own evidence. Don't slide from one to the other.
One vivid story that X led to Y is evidence that this can happen; it is not evidence of how often it happens. Probability claims require base-rate data, not memorable cases.
Arguments often start with evidence of "some cases" and smuggle in a universal conclusion. "Some investments harm people" → "Investing is wrong" is a hidden leap. The gap is the missing universal premise, and you should name it.
GLOSSARYKey terms introduced
- Empirical premise
- A claim about the world that could be verified or falsified by observation and evidence.
- Universal claim
- A claim of the form "Every X is Y" or "Whenever X, Y" — must hold in all cases.
- Probabilistic claim
- A claim about tendencies or frequencies — "X often leads to Y." Requires statistical evidence.
- Counterexample
- A specific case that satisfies the condition of a universal claim but falsifies its conclusion, thereby refuting the universal.
- Anecdotal evidence
- Evidence drawn from one or a few individual cases; supports existential claims but not universal or probabilistic ones.
- Representative sample
- A sample drawn such that its composition reflects the population it is meant to describe.
- Alternative explanation
- A competing account that explains the same evidence without endorsing the original causal claim.
- Quantifier
- A word that specifies scope — e.g., "all," "some," "never," "tends to." Determines the evidence standard a claim must meet.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1Why does a single counterexample defeat a universal claim but not a probabilistic claim?›
A universal claim ("every X is Y") requires the relationship to hold in all cases with no exceptions. One case where X is not Y is therefore sufficient to show the claim is false. A probabilistic claim ("X tends to be Y") only requires the relationship to hold for some proportion of cases — one exception is compatible with it still being true in 90% of cases, and refuting it requires systematic evidence about the proportion, not a single outlier.
2In the running example, what exactly does the counterexample — a productive investment that creates new value — prove about P2?›
It proves that the universal form of P2 is false: it is not the case that every investment profit requires an equivalent loss elsewhere. It does not prove that investment profits never harm anyone — only that the "whenever" quantifier is wrong. The narrower question of how often investment profits come at others' expense remains open.
3What distinguishes anecdotal evidence from systematic evidence, and what kind of claim can each legitimately support?›
Anecdotal evidence is drawn from one or a few specific cases — vivid, concrete, and individually compelling. Systematic evidence is drawn from a large, representative sample using a consistent method. Anecdotes can legitimately support existential claims (showing that something is possible or has occurred) and can generate hypotheses. They cannot support universal or high-probability claims. Systematic evidence is needed for those.
4Why might two people look at the same evidence — "some investment gains are at others' expense" — and reach very different conclusions? What reasoning tool resolves this?›
One person might read "some cases" as supporting a universal ("investment is always zero-sum"), while another reads the same evidence as compatible with a probabilistic claim ("it sometimes is"). The resolution is careful attention to quantifiers: identify exactly what the evidence establishes (an existential claim — these cases exist), then separately evaluate what proportion of all investment is zero-sum (a probabilistic question requiring base-rate data). The evidence about "some cases" does not, by itself, answer the proportional question.