Skip to sample
Synlecto / Public samples

Sample Space

John Tsitsiklis · MIT OpenCourseWare

About this guide and its source

Learn to describe a random experiment before assigning probabilities.

These curated examples are not untouched extension output. Reviewed against the public lecture; use the source as the authority.

5 concepts6 flashcards4 quiz questions1 mission

Before you start: Recognise a set as a collection of distinct elements. No probability calculations are needed.

Covers the complete short lecture: the two modelling steps, outcomes, exclusivity, exhaustiveness and useful detail. It does not teach how to assign a probability law. Practice scenarios are newly written Synlecto exercises.

Links return to source sections, not exact sentence timestamps. Editorial edition: 2026-11-09.

Original MIT lecture (new tab) · Official transcript (new tab)

Try this curated guide without signing in. Practice resets when you leave or reload. Nothing is saved to an account.

Chapter 01

Starting points

Build the model in this order so that likelihood does not distract from what is possible.

Name the experiment

Say what is done once and what is recorded at the end. In the lecture’s coin model, record heads or tails. If you also record rain or no rain, the recorded result is a face/weather pair. Changing what you record changes the distinctions the model must preserve.

Use the exactly-one check

For any completed experiment, you should be able to select one and only one outcome. No match means a gap; two matches mean overlap.

Leave likelihood for the next step

Only after defining the outcomes do you specify a probability law.

Chapter 02

Concept map

Lecture concept map

Explore the ideas, then return to the lecture to check the detail and context.

Use the up and down arrow keys to move between concepts. Press Home or End to jump to the first or last concept.

Concept 1 of 5

Focused concept: Outcomes before probabilities. Concept 1 of 5.

Focused concept

Outcomes before probabilities

Concept 1 of 5

Definition

A probabilistic model first specifies what can happen. It then assigns a probability law to express beliefs about how likely those outcomes are.

Example

For a coin experiment, identify heads and tails before deciding what probability to assign to either.

Common confusion

Listing outcomes already tells us their probabilities.

Check the distinction

A list of possibilities and a probability law answer different questions. Two listed outcomes do not by themselves establish equal chances.

Chapter 03

Compare

These checks catch different modelling errors.

Exclusive asks: can two descriptions both apply?

This is the at-most-one test. Synlecto example: for a die with faces 1 to 6, the descriptions “even” and “greater than 4” both match 6. Different labels have not produced mutually exclusive outcomes. Finding one such result is enough to show the problem.

Exhaustive asks: could no description apply?

This is the at-least-one test. Synlecto example: {1, 2, 3, 4, 5} misses a possible roll of 6. The listed faces are distinct, so exclusivity passes, but exhaustiveness fails. Passing one condition never substitutes for checking the other.

Validity and usefulness are different

The four coin/weather combinations can be valid because every result matches one complete pair. For a coin-face-only question, the simpler space preserves everything you need. For a weather comparison, it discards information you need. More detail is useful only in relation to the question, not because a longer list is automatically better.

Chapter 04

Process

A modelling checklist distilled from the lecture.

Describe, test, simplify

A sample space is a modelling choice, not just a list of words. Work from the experiment to the observation you need to preserve. The two conditions then provide a practical way to test your proposed outcomes.

  1. Specify the experiment and observation

    Say what is done once and what is recorded at the end. Recording only a coin’s face is different from recording its face together with the weather.

  2. Try concrete completed experiments

    Check whether each possible result fits at least one description and no more than one. No matching description reveals a gap; two matching descriptions reveal overlap.

  3. Remove irrelevant distinctions

    Two spaces can both be valid. Prefer the simpler description when the additional information cannot affect the question you want the model to answer. This choice comes before assigning probabilities.

Try it: can a list be exhaustive but fail exclusivity?

Yes. In a Synlecto die example, “odd”, “even” and “six” cover all six faces, but a roll of 6 fits both “even” and “six”. The gap check passes while the overlap check fails.

Then assign probabilities

A valid outcome set completes the first modelling step, not the whole probability model. Nothing about listing two or four outcomes establishes equal chances. Keep “what could happen?” separate from “how likely is it?” until a probability law is supplied.

Chapter 05

Applied examples

The lecturer’s coin example shows why modelling choices matter.

Coin face only

Use {heads, tails} when the observation of interest is just the face of the coin.

Try it: after observing tails, is the sample space just {tails}?

No. Tails is the realised outcome. The sample space still describes both possible results before this one experiment was carried out: {heads, tails}.

Coin face and weather

The lecture’s richer model records two observations in one outcome: coin face and whether it is raining. Its four pairs are (heads, rain), (heads, no rain), (tails, rain) and (tails, no rain). A pair is one complete outcome, not two competing entries.

  1. Classify a completed experiment

    A tails result on a rainy day belongs to (tails, rain). It does not also belong to (heads, rain), because the complete pair must match.

  2. Check the whole model

    Each of the two faces has a rain and a no-rain case. Under the lecture’s model, every completed experiment therefore has one matching pair.

  3. Choose the question before choosing detail

    The richer space retains information needed to investigate a possible weather relationship. The coin-only space discards that information, which is appropriate if only the face matters. Neither choice proves that weather affects the coin.

Try it: does four outcomes mean a probability of one quarter for each?

No. The sample space describes possibilities. A probability law is a separate modelling step, and equal likelihood has not been established.

A hypothesis is not a conclusion

The suggestion that weather might matter motivates a richer model. The example does not establish that weather changes a coin’s behaviour.

Repair a model by testing actual results

Synlecto worked example: a die has faces 1 to 6, and only parity matters. Someone proposes {odd, even, six}. Every face has a description, but the proposal still fails.

  1. Find the witness to the error

    A result of 6 matches both “even” and “six”. This is overlap, not a missing possibility: exclusivity fails while exhaustiveness passes.

  2. Repair for the stated question

    Use {odd, even}. The odd result groups {1, 3, 5}; the even result groups {2, 4, 6}. Each face matches exactly one recorded category, and parity is preserved.

  3. Check what the repair discards

    The grouped model cannot distinguish 2 from 6. That is acceptable for a parity-only question. If the exact face matters, use {1, 2, 3, 4, 5, 6} instead. Neither model by itself assigns probabilities.

Try it: is deleting “even” from the original proposal another valid repair?

No. The remaining {odd, six} leaves 2 and 4 without a description. Removing an overlap has introduced a gap. Retest both conditions after any repair.

Chapter 06

Common confusions

Common traps, with the distinction to keep in mind.

Outcomes before probabilities

Common confusion: Listing outcomes already tells us their probabilities.

Check the distinction: A list of possibilities and a probability law answer different questions. Two listed outcomes do not by themselves establish equal chances.

The sample space

Common confusion: The sample space is the outcome that actually happened.

Check the distinction: The realised outcome is one element of the sample space. The space includes the other possibilities too.

Mutually exclusive outcomes

Common confusion: Different names are enough to make outcomes distinct.

Check the distinction: Check what each description covers. Overlapping descriptions can still both be true.

Collectively exhaustive outcomes

Common confusion: A list is acceptable if each listed outcome is possible.

Check the distinction: It must also include all modelled possibilities. Omitting an outcome leaves some experiments unclassified.

Choose detail for the question

Common confusion: The most detailed sample space is always the correct one.

Check the distinction: Both descriptions can be legitimate. Prefer the simpler one when the additional distinction is irrelevant.

Chapter 07

Decision points

Ask what the model must let you distinguish.

Does weather belong in this model?

Keep it when investigating a question about coin face and weather together. Omit it for a question that only needs the coin face.

Does a possible case fit twice or nowhere?

Check the descriptions before doing arithmetic. If a case fits twice, refine overlapping descriptions; if it fits nowhere, add the missing possibility or revise the stated model. Assigning probabilities cannot fix an outcome list that does not classify the experiment unambiguously.

Chapter 08

Checklist

Close the guide and explain these in your own words.

Define Ω

Distinguish the entire sample space from a single realised outcome.

Diagnose two errors

Give one example of overlapping descriptions and one of a missing outcome. Name the failed condition in each.

Defend a level of detail

Explain why the coin-only and coin/weather spaces can both be valid, and choose between them for a stated question.

When you come back

Try these from memory in a later session. These are suggested prompts, not a scheduled reminder.

Without looking back, design a sample space for one coin flip when only its face matters. Then add weather as an observation.

Check your explanation

The first space has heads and tails. The second has four face/weather pairs; every experiment matches one pair.

Explain why a model can be exhaustive but not exclusive.

Check your explanation

Covering every possibility does not prevent overlap. For a six-faced die, {odd, even, six} covers all faces, but 6 fits twice. Repair the list and test all six faces again. A definition without a witness does not yet demonstrate that you can diagnose the error.

Try the flashcards
Sample Space · Curated sample | Synlecto