Induction

The conclusion goes beyond what was observed

A record describes observed cases. An inductive conclusion reaches further: perhaps to the next event, a larger group or an expectation about a pattern. The evidence can be useful without making every contrary outcome impossible. To evaluate the reasoning, identify exactly which step moves beyond the record and what supports that step.

This does not mean that induction is merely guessing. Evidence can make some conclusions better supported than others. But the support needs an explanation: relevant cases, reliable observations and a conclusion with an appropriate scope. Neither a confident speaker nor the word 'probably' supplies those connections on its own.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy

An invented bus record separates frequency from prediction

Imagine a notebook containing eight weekday morning bus arrivals on one route. Six were on time. If the record is accurate, six of eight, or seventy-five percent, describes those recorded arrivals. It is a statement about the notebook's eight cases. The conclusion that the next bus will probably be on time reaches beyond them.

To assess that prediction, ask which trip the next bus is making. Is it on the same route and at a comparable time? Did the timetable change? Were late arrivals recorded as carefully as early ones? Eight morning observations do not automatically represent evening trips, weekends or every route. The numerical fraction does not by itself establish a seventy-five-percent probability for any of those targets.

The relevant conclusion could be much narrower than 'buses are reliable'. Someone might say that this small record gives a limited reason to expect a similar weekday morning arrival to be on time, while further information could change that expectation. The qualification explains how far the evidence is being used. It does not turn the notebook into a complete probability model.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy

A larger sample can still have a selection problem

Suppose a fictional reading club asks for feedback only through its most active online group. Many members of that group respond enthusiastically. The responses establish what the respondents reported, assuming they were recorded accurately. They do not automatically establish that every club member shares the same view. People who never use the group had no comparable opportunity to respond.

Collecting more replies from the same selected channel may make that channel's views clearer while leaving the coverage problem in place. Ask how the sample relates to the target group. A relevant sampling procedure and reliable records can improve the basis for generalization; sheer size is not a substitute for explaining who was included and who was absent.

Voluntary feedback can still be valuable for a limited purpose, such as learning what participating members liked or disliked. The mistake is using it for a different conclusion without acknowledging the change. Specify whether you are describing respondents, group users or the whole club. Those are different targets.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy

Missing cases should stay missing in the argument

Return to the bus notebook. Perhaps the observer left before the bus arrived on two other mornings, so there is no arrival time for those trips. A blank entry is not an on-time arrival, and it is not automatically a late one either. It marks an observation that was not completed. Treating those blanks as successful trips would give the record a precision it does not have.

The missingness may also be relevant. If observation stopped whenever waiting became inconvenient, long waits could be underrepresented. That is a possible problem to investigate, rather than a fact already proved about the notebook. Good evaluation distinguishes the recorded cases, the missing cases and the reasons missingness might affect the conclusion.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy

New evidence can change the strength of an expectation

Suppose the timetable changes after the eight observations. The old record still describes the earlier trips, but its relevance to the new schedule now needs examination. Revising the expectation would not make the earlier observations false. It would acknowledge that the connection from those observations to the new target has changed.

Likewise, an exception can matter in different ways. One late trip falsifies the universal claim that every recorded trip was on time, but that was never true of the six-out-of-eight notebook. It need not falsify a carefully qualified expectation. Ask whether the original conclusion allowed exceptions before deciding what the new case defeats.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy

Predictions, universal rules and explanations have different tasks

An induction need not end with an 'all' statement. The prediction about the next bus concerns one particular future case. The general-to-specific versus specific-to-general slogan therefore misses some everyday reasoning. Compare the support for the conclusion, not only the grammatical shape of its sentences.

A proposal about why a bus was late poses another question: which explanation fits the available evidence? That is different from estimating a pattern of arrivals. Ordinary investigations can combine explanation, prediction and deductive consequences of stated rules. Dividing the investigation into its component claims helps you evaluate each connection rather than giving the whole task one reassuring label.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy

The practical evidence check leaves a deeper philosophical question

Improving a sample does not turn an evidence-based expectation into a deductive guarantee. The deeper philosophical issue concerns what justifies moving from observed experience to unobserved cases. The separate Hume reading guide follows that question in detail; it is useful after you can identify the step in an ordinary example.

For the present argument, write the actual record, the target conclusion and the assumptions connecting them. Explain what could strengthen or weaken that connection. A verdict such as 'the responses are useful for active group users but do not yet establish the views of the whole club' identifies a concrete limitation. It is more informative than declaring that all induction is good or that all uncertainty makes reasoning useless.

Further reading: Timothy Shanahan, Internet Encyclopedia of Philosophy