If the evidence is true, does the conclusion follow?
Information judgment · Lesson 5: Check the inference
Estimated time: 20 minutes. Suggested preparation: lessons one to four.
Teaching note: All cases are fictional. Temporarily accept the observations supplied and examine what they support. They are not real product or research findings.
What you will learn
- Separate observations, explanations and recommendations.
- Propose testable alternative explanations.
- Use appropriate counterexamples to examine universal claims.
- Find omitted options and rewrite conclusions that exceed the evidence.
1. Accept the evidence, then ask what follows
A company finds that employees using a new tool have higher average output. Therefore, the tool makes employees more efficient, and every team should buy it immediately.
Even if the first statement is accurately measured, the next two steps require support. Identify the observation, causal explanation and recommendation, then a question that could change the purchasing decision.
This lesson does not evade reasoning by doubting every number. Accept the observation provisionally, then examine assumptions added between observation, explanation and action.
2. One observation can have several explanations
Tool users may produce more because of the tool, because already productive employees are more likely to adopt it, or because users receive different tasks. Several factors may operate together.
Proposing an alternative does not establish it as true. It shows that the observation alone has not distinguished the possibilities.
A useful alternative specifies a process and a checkable expectation. “Already productive staff adopt the tool” suggests inspecting output before adoption. Merely saying “unknown factors may exist” does not finish the analysis.
If differences predate adoption, the entire later gap cannot simply be credited to the tool. That does not prove the tool contributes nothing. Comparable tasks, baseline records and suitable comparison designs can help, subject to implementation and other limitations.
3. How far is it from a case to everyone?
“These three people learned it” establishes successful cases, not that anyone can learn it in the same time. Consider starting skills, resources, practice time and unsuccessful participants.
For a strict claim that all X are Y, one genuine X that is not Y refutes that universal claim. One failure does not refute “most succeed” or “it helps on average”; those require appropriate data.
A counterexample must meet the original conditions. If a course specifically targets experienced learners, a complete beginner's experience does not directly refute that qualified promise.
4. What lies outside the two offered options?
“Adopt it everywhere now or fall behind forever” may omit pilots, staged adoption, alternative tools or process changes.
Do not reject a claim merely because it says “either.” Some precisely defined rules have two outcomes, such as whether a submission occurred before a specified deadline. Moving from two states to only two possible actions still requires examination.
Find a feasible third option and state its constraints. An impossible fantasy does not establish a flaw in the decision frame.
5. Match conclusion strength to evidence
| Available material | Supported wording | What does not directly follow |
|---|---|---|
| One successful case | At least one case succeeded under these conditions | Everyone will succeed |
| Observed differences between groups | These groups differed in this observation | A tool caused the entire difference |
| Insufficient material to assess an effect | The effect remains unresolved; these checks are needed | Ineffectiveness has been proven |
| Support for an intervention in specific conditions | There is support under those conditions | Adoption is worthwhile everywhere |
Insufficient evidence need not freeze every decision. Consider costs, risks, reversibility and timing. A limited pilot with predefined measures and stopping conditions may be preferable to immediate full adoption.
Absence of evidence can sometimes be informative: a reliable check capable of detecting an expected effect that finds none should update judgment. Examine the check's scope and detection capability. Do not always equate “not found” with “impossible,” or permanently dismiss negative findings.
6. Exercise one: what does the comparison rule out?
Fictional observations: voluntary tool users in Group A averaged 12 items per person per week before adoption and 15 afterward. Non-users in Group B went from 8 to 11 over the same period. Changes in task allocation and staffing are unknown.
Assess “A produced four more items afterward, so the tool adds four items per person per week.” Propose an alternative explanation and a further check.
Explore the answer and explanation
Explanation
A already exceeded B by four items before adoption; the gap remained four afterward. Both increased by three. The later four-item difference cannot by itself be attributed to the tool.
An alternative is a pre-existing output difference combined with a shared change that raised both groups. Check task difficulty, staffing, shared process changes and longer baseline trends.
The difference between changes is zero as a descriptive calculation. Without comparability and other necessary conditions, it does not establish an exactly zero causal effect. It weakens this argument without completing a causal evaluation.
7. Exercise two: is one counterexample enough?
Compare fictional promises:
A. Every enrollee can finish a project within one week.
B. Under the stated conditions, most enrollees can finish a project within one week.
One enrollee is confirmed not to have finished in a week. What does that imply for each claim? If A explicitly applies only to people practicing two hours daily, what else needs checking?
Explore the answer and explanation
Explanation
For A without further conditions, the enrollee is a counterexample to “every.” For B, one non-completer does not establish whether most completed. Total counts, completion counts and conditions are needed.
If A explicitly requires two hours of daily practice, verify that this person met the condition. Do not invent new restrictions after a counterexample appears merely to shield the original promise.
A better statement does not arbitrarily replace “every” with “most”: “At least one enrollee did not finish within a week; the overall completion proportion still needs checking.”
8. Exercise three: from explanation to decision
Fictional workplace message:
Our competitor delivers faster after adopting automation. We must buy the same system today or be eliminated.
Identify the observation, implied cause and recommendation. Propose a feasible third option and write a bounded recommendation with pilot acceptance criteria.
Explore the answer and explanation
Explanation
The observation is faster delivery after adoption; the implied cause is the system; the recommendation is immediate purchase, with elimination presented as the only alternative. Other changes have not been excluded, and suitability for this team has not been established.
One option is a limited trial on repetitive tasks while retaining the current process. Check trial availability, budget and timing; if a pilot is infeasible, choose another realistic option.
A recommendation: “Check which tasks improved and under what implementation conditions, then evaluate representative tasks from our team. Record processing time, errors, human review and total costs, with acceptance and stopping criteria defined before deciding whether to expand.”
Illustrative criteria might require maintained quality, lower total time including review, and costs within an agreed budget. These are example decision criteria, not universal effectiveness guarantees.
9. Review your reasoning
Assess how strong a conclusion the argument supports, not whether you must favor or oppose a particular plan.
10. An inference review
The next lesson will combine claims, emotion, sources, numbers and reasoning in a complete information review. Continue to the integrated exercise using the navigation below.
Entries stay in this browser, separately for each language. Export a copy for backup.