When Is “Uniform” a Valid Assumption?
A uniform model is tempting when a random variable has a short list of possible values. But having several possible values does not make them equally likely. The key question is not how many values appear on the list; it is whether the probability assigned to each value is the same under the stated random mechanism.
This distinction matters whenever a variable is produced by mapping more detailed outcomes into a smaller set of values. If some values can be produced in more ways than others, they may have larger probabilities even when the underlying outcomes are equally likely. A related trap is treating a continuous uniform variable as though each point in its interval had positive, equal probability.
A discrete uniform model on a finite set describes equal probability for each value of the variable. It does not require the values to be consecutive integers. Conversely, a process can have equally likely elementary outcomes without giving equal probabilities to the values of a variable derived from those outcomes. Check the mapping from outcomes to values before assigning a uniform probability.
Underlying Outcomes and Unequal Numbers of Ways
Suppose an experiment has \(N\) equally likely elementary outcomes, and a random variable \(X\) assigns a value to each outcome. For any possible value \(x\), let \(m_x\) be the number of elementary outcomes that produce \(x\). Then
This formula shows the condition behind a discrete uniform model: the values must have equal numbers of equally likely outcomes mapped to them (or otherwise be shown to have equal probabilities). Equal likelihood of the underlying outcomes alone is not enough.
Worked Examples
Worked Example: A Sum of Fair Dice Is Not Uniform
Two fair, six-sided dice are rolled. Assume each die has faces 1 through 6, each face is equally likely, and the dice are independent. Let \(X\) be their sum. Find \(P(X=8)\), \(P(X=2)\), and assess whether \(X\) is discrete uniform on the integers from 2 through 12. There is no interest or mortality assumption in this probability model.
Method. Count ordered pairs of die results. There are \(6\cdot 6=36\) equally likely ordered pairs. The two results are ordered because, for example, \((2,6)\) and \((6,2)\) are distinct outcomes of the roll.
To obtain a sum of 8, the ordered pairs are \((2,6),(3,5),(4,4),(5,3),(6,2)\). Thus \(m_8=5\), and
A sum of 2 occurs only for \((1,1)\), so \(m_2=1\) and
The support contains 11 possible sums, from 2 through 12. If \(X\) were discrete uniform on this support, each sum would have probability \(1/11\). But \(1/36\ne 1/11\), so \(X\) is not discrete uniform. In fact, the different sums have different numbers of ordered pairs producing them.
Check. The probabilities found are consistent with the 36 equally likely outcomes: one outcome gives a sum of 2, while five give a sum of 8. The mistake to avoid is assigning \(1/11\) to every possible sum just because there are 11 possible sums.
Worked Example: Weighted Outcomes Require Weighted Probabilities
A hypothetical claim-triage system assigns a randomly selected claim to one of three review levels. Level 1 is assigned with probability \(0.50\), Level 2 with probability \(0.30\), and Level 3 with probability \(0.20\). Let \(X\) be the level number. Find \(P(X\ge 2)\), \(E[X]\), and \(P(X=3\mid X\ge 2)\). Treat the stated probabilities as the complete model; there is no interest or mortality assumption.
Method. Use the given probability mass function. The support is \(\{1,2,3\}\), but the probabilities are not equal, so \(X\) is not discrete uniform.
The event \(X\ge 2\) consists of Levels 2 and 3. Therefore,
For a discrete random variable, \(E[X]\) is the sum of each value multiplied by its probability:
For the conditional probability, both the numerator and denominator refer to the relevant event:
Check. The probabilities sum to \(0.50+0.30+0.20=1.00\), and the conditional probability is between 0 and 1. A common wrong calculation is to replace each given probability by \(1/3\), producing \(P(X\ge 2)=2/3\). That calculation silently changes the model; the three levels were not stated to be equally likely.
Worked Example: Discrete and Continuous Uniform Probabilities
Compare two models. In the first, \(D\) is selected uniformly from the 13 integers \(0,1,\ldots,12\). In the second, \(U\) has a continuous uniform distribution on the interval \([0,12]\), with density \(f_U(u)=1/12\) for \(0\le u\le 12\). Find \(P(3\le D<8)\), \(P(3\le U<8)\), \(P(D=5)\), and \(P(U=5)\). These are mathematical probability models; no interest or mortality assumption applies.
Discrete model. The integers satisfying \(3\le D<8\) are \(3,4,5,6,7\): five values out of 13 equally likely integers. Hence
Each integer has probability \(1/13\), so \(P(D=5)=1/13\approx 0.0769\).
Continuous model. For a continuous uniform variable on \([0,12]\), the probability of an interval is its length divided by the total interval length. The interval from 3 up to, but not including, 8 has length \(8-3=5\), while the full support has length 12. Thus
A single point has length zero, so \(P(U=5)=0\). Including or excluding an endpoint does not change the probability for this continuous model, because a single point has probability zero.
Check. The discrete calculation counts five included integers; the continuous calculation uses an interval length of five. The denominators differ because one model has 13 equally likely points and the other spreads probability across an interval of length 12. The mistake to avoid is treating the continuous model as if each point had probability \(1/12\).
Why the Counting Rule Holds
The rule \(P(X=x)=m_x/N\) follows directly from the definition of equally likely elementary outcomes. Let the experiment have \(N\) such outcomes, each with probability \(1/N\). The event \(X=x\) consists of exactly the \(m_x\) outcomes assigned the value \(x\). Since those elementary outcomes are mutually exclusive, their probabilities add:
Therefore, values of \(X\) are equally likely exactly when their outcome counts \(m_x\) are equal in this setting. This gives a quick diagnostic: count how many elementary outcomes lead to each value. If those counts vary, the induced variable is not discrete uniform.
Continuous Uniform Models: Density Is Not Point Probability
A continuous uniform variable on \([a,b]\), where \(a<b\), has constant density \(1/(b-a)\) throughout that interval and density zero outside it. For \(a\le c\le d\le b\), its interval probability is
The density is a rate of probability per unit of the variable, not the probability of an individual point. In particular, for any \(u\) in the interval, \(P(U=u)=0\). A continuous uniform model is not a discrete uniform model with an enormous number of values; the sample space and the probability calculation are different.
A discrete uniform variable on consecutive integers \(a,\ldots,b\) instead assigns probability \(1/(b-a+1)\) to each integer. The “plus one” counts the endpoints as included values. In a continuous interval, the length is \(b-a\), not \(b-a+1\).
| Feature | Discrete uniform on \(n\) values | Continuous uniform on \([a,b]\) |
|---|---|---|
| Probability assigned | Each listed value has probability \(1/n\) | Probability is distributed over intervals by length |
| Point probability | A support value has probability \(1/n\) | Every individual point has probability 0 |
| Interval calculation | Count included support values and divide by \(n\) | Divide interval length by \(b-a\) |
| Endpoint handling | Included integer endpoints count as values | Adding or removing one endpoint does not change probability |
Common Modeling Traps
- Making possible values equally likely by assumption. A list of possible values defines a support, not a probability mass function. Use the random mechanism or stated probabilities to establish likelihood.
- Confusing equal elementary outcomes with equal values. A value produced by many elementary outcomes can have greater probability than a value produced by only a few.
- Using the wrong denominator for consecutive integers. There are \(b-a+1\) integers from \(a\) through \(b\), inclusive. The length \(b-a\) is not the number of discrete values.
- Assigning a continuous point a positive probability. A density value is not a point probability. Calculate continuous probabilities over intervals.
- Ignoring endpoint conventions. For a discrete variable, changing an endpoint can add or remove a possible value. For a continuous uniform variable, changing inclusion of a single endpoint does not change the interval probability.
These distinctions help prevent an error at the first step of a calculation: choosing the wrong model. Once the model is correct, the relevant calculation is usually direct—sum the specified discrete probabilities, count equally likely discrete values when justified, or use an interval-length ratio for a continuous uniform variable.
Check Your Understanding
For each problem, identify the model before calculating.
- A fair eight-sided die has faces numbered 1 through 8. Find the probability of an even result and the probability of a result at least 6. (Answer: \(1/2\) and \(3/8\))
- A hypothetical service assigns a randomly reviewed file to category A, B, or C with probabilities \(0.40, 0.35,\) and \(0.25\), respectively. Find the probability of category B or C and the expected category number if A, B, C are coded 1, 2, 3. (Answer: \(0.60\) and \(1.85\))
- Let \(D\) be discrete uniform on the integers from 4 through 15, inclusive. Find \(P(7\le D\le 10)\) and \(E[D]\). (Answer: \(1/3\) and \(9.5\))
- Let \(U\) be continuous uniform on \([2,14]\). Find \(P(5<U\le 11)\) and \(P(U=7)\). (Answer: \(0.5000\) and \(0\))
- Two fair, independent six-sided dice are rolled. Find the probability their sum is at least 10, and explain why treating sums 2 through 12 as equally likely gives the wrong result. (Answer: \(1/6\); sums 10, 11, and 12 have 3, 2, and 1 ordered outcomes, respectively, out of 36)