Uniform Probability Models and Symmetry

When Is It Reasonable to Treat Outcomes Equally?

A uniform probability model assigns the same probability to every individual outcome in a finite sample space. It is a simple and useful model, but equal probabilities are not automatic: they must be supported by the way the experiment is modeled. A fair die is a familiar example because its faces are treated symmetrically by the rolling mechanism. A loaded die may have the same six possible faces without making those faces equally likely.

The sample space matters, too. For a single card drawn from a well-shuffled deck, the outcomes can be the 52 individual cards. The four suits are categories made up of those outcomes; they are not the individual outcomes themselves. This distinction lets us state precisely what is assumed to be equally likely.

Definition: Let \(\Omega\) be a nonempty finite sample space. The uniform probability law on \(\Omega\) assigns probability \(P(\{\omega\})=1/|\Omega|\) to each outcome \(\omega\in\Omega\), where \(|\Omega|\) is the number of outcomes in \(\Omega\). For any event \(A\subseteq\Omega\), its probability is the sum of the probabilities of the outcomes in \(A\).

This definition specifies a model, not a physical guarantee. To use it for an experiment, we need a reason to regard the modeled outcomes as equally likely—for example, a design that treats the outcomes symmetrically, or an explicit idealization such as a fair die.

Event Probabilities Under a Uniform Law

Because an event in a finite sample space is a finite collection of outcomes, its probability is the sum of the point probabilities of those outcomes. Under the uniform law, every term in that sum is \(1/|\Omega|\).

Theorem: If \(\Omega\) is a nonempty finite sample space with the uniform law and \(A\subseteq\Omega\), then \(P(A)=|A|/|\Omega|\).

Proof. The event \(A\) consists of \(|A|\) distinct outcomes. By finite additivity, its probability is the sum of the point probabilities over those outcomes. Each point probability is \(1/|\Omega|\), so

P(A)=\sum_{\omega\in A}P(\{\omega\}) =\sum_{\omega\in A}\frac{1}{|\Omega|} =\frac{|A|}{|\Omega|}.

If \(A\) is empty, the sum is zero and the same formula gives \(P(A)=0\). If \(A=\Omega\), it gives \(P(A)=1\), as required. \(\square\)

The theorem is a consequence of the uniform model; it does not decide whether that model is appropriate. In particular, the formula cannot be applied just because a list of outcomes is finite. Equal point probabilities must first be justified.

Worked Example: A Fair Six-Sided Die

Model one roll of a fair six-sided die by \(\Omega=\{1,2,3,4,5,6\}\), where each outcome records the face showing. The fair-die assumption says that these six faces have equal probability. Since their probabilities sum to one, each is \(1/6\).

Let \(A\) be the event of rolling an even number. Then \(A=\{2,4,6\}\), so \(|A|=3\) and \(|\Omega|=6\). Therefore, \(P(A)=3/6=1/2\).

This conclusion concerns a model of a fair die. If the die is biased, the outcome set stays the same, but the equal-probability assumption may fail.

Symmetry as a Justification

Symmetry can justify equal point probabilities when it is a symmetry of the probability model, not merely a resemblance in a drawing. A precise version uses transformations that permute the outcomes while preserving the model. If such transformations can carry any outcome to any other, then all outcomes must have the same probability.

Theorem: Let \(\Omega\) be finite, and suppose a collection of bijections \(g:\Omega\to\Omega\) preserves the probability law: for every such \(g\) and every event \(A\subseteq\Omega\), \(P(gA)=P(A)\), where \(gA=\{g\omega:\omega\in A\}\). If for every \(\omega,\eta\in\Omega\) at least one of these bijections satisfies \(g\omega=\eta\), then the law is uniform on \(\Omega\).

Proof. Fix any two outcomes \(\omega,\eta\in\Omega\). By the stated condition, there is a probability-preserving bijection \(g\) with \(g\omega=\eta\). Applying invariance to the singleton event \(\{\omega\}\) gives \(P(\{\eta\})=P(g\{\omega\})=P(\{\omega\})\). Thus every point probability has a common value \(c\). The singleton events partition \(\Omega\), so finite additivity gives \(1=P(\Omega)=\sum_{\omega\in\Omega}P(\{\omega\})=|\Omega|c\). Hence \(c=1/|\Omega|\), which is the uniform law. \(\square\)

For a die modeled as fair, the symmetry that permutes the faces is part of the idealized physical model: no face is given a different chance by the rolling procedure. The theorem explains why that assumption leads to equal probabilities. It does not show that any particular die is fair; fairness is a modeling assumption that may be assessed using knowledge of the die and how it is rolled.

Worked Example: One Card From a Well-Shuffled Deck

Suppose a standard deck has 52 distinct cards and is well shuffled, with each card equally likely to be the top card. Use \(\Omega\) equal to the set of individual cards. The model assigns \(1/52\) to each card.

Let \(A\) be the event that the top card is a heart. There are 13 hearts among the 52 cards, so \(P(A)=13/52=1/4\).

For the event that the top card is the ace of hearts, there is just one qualifying card. Thus \(P(\{\text{ace of hearts}\})=1/52\).

The equal-probability assumption here is about individual cards. Equal probabilities for the four suits follow because each suit contains 13 cards; they are not a separate assumption that every possible group of cards has the same probability.

The Sample Space Must Match the Question

Two descriptions of the same experiment can use different outcome spaces. A uniform law on individual objects need not make every category or summary value equally likely. Probabilities for those events are obtained by adding the point probabilities of the individual outcomes they contain.

Worked Example: Card Identities Versus Card Ranks

Use the same single-card model, with all 52 individual cards equally likely. Let \(R\) be the event that the card has rank 7. There are four cards of rank 7, one in each suit, so \(P(R)=4/52=1/13\).

The 13 possible ranks are not 13 equally likely outcomes under this model: each rank corresponds to four equally likely card identities. For example, the event “rank is an ace” also has probability \(4/52=1/13\), while the event “the card is a heart” has probability \(13/52=1/4\).

If the question is about rank rather than card identity, we can use ranks as the outcome descriptions, but their probabilities must still reflect the underlying card-draw model. Choosing a coarser sample space does not make its outcomes equally likely.

When Equal Probabilities Are Not Justified

Having a finite list of possible outcomes does not imply a uniform law. Consider a device that outputs \(a\), \(b\), or \(c\), with probabilities \(1/4\), \(1/4\), and \(1/2\), respectively. These probabilities are nonnegative and sum to one, so they define a probability law, but it is not uniform. The fact that there are three listed outcomes does not alter the assigned probabilities.

Likewise, visual balance alone may be misleading. A die can look regular but be weighted; a shuffle may not be sufficiently random; and outcomes defined at different levels of detail may represent different numbers of underlying possibilities. The right question is not simply whether outcomes look alike, but whether the experiment and the chosen model treat them alike.

Common pitfall: “There are six possible results, so each has probability \(1/6\).” The number of outcomes alone does not establish equal probabilities. The correction is to state the probability assumption—such as that the die is fair—and explain why it is appropriate for the experiment. If the outcomes are not equally likely, assign or derive their probabilities from the model instead.

Uniform models are useful because event probabilities reduce to comparing the number of outcomes in an event with the total number of outcomes. But this convenience comes after the modeling decision, not before it. First identify the outcomes relevant to the question; then decide whether the experimental mechanism makes those outcomes equally likely; only then use the uniform-law formula.

Check Your Understanding

Answer each question by identifying the sample space and the probability assumption being used.

A die has faces numbered 1 through 6 and is assumed fair. What is the probability of an outcome greater than 4?
For a well-shuffled standard deck, why does “the card is a queen” have probability \(4/52\), rather than \(1/13\) by an assumption of equal card identities?
A device has three possible outputs with probabilities \(0.2\), \(0.3\), and \(0.5\). Is its law uniform? Explain.
State the symmetry condition in the theorem that guarantees all point probabilities are equal.
Why does a finite sample space, by itself, not justify assigning equal probability to every outcome?