Statistics › Section D · Discrete random variables › Topic D.2
Expected value and standard deviation
A distribution lists everything a random variable can do; two numbers summarise it. The expected value is its balance point, the standard deviation its typical spread - together they underpin how games, insurance premiums and payouts are priced.
The distribution's centre and spread
A distribution lists everything a random variable can do; two numbers summarise it. The expected value is the balance point - the long-run average of the values, weighted by their probabilities. The standard deviation measures how far, typically, the values sit from that centre. Together they do for random variables what the mean and SD did for data, and they underpin pricing in insurance, gambling and game design.
Expected value
The expected value of \(X\) is the probability-weighted average of its values:
Each value is multiplied by its probability, and the products add. It is the long-run average: repeat the process many times and the running mean of the results closes in on \(\mu\). The expected value need not be a possible value - the next card's game averages to a loss even though no single play loses exactly that amount, just as a country can average 1.4 children per household while no household has 1.4 children.
Worked example: is the game fair?
You pay $3 for one die roll and win $12 on a six, nothing otherwise. The net gain \(X\) is +9 with probability \(\frac{1}{6}\) and -3 with probability \(\frac{5}{6}\). 1. \(E(X) = 9 \times \frac{1}{6} + (-3) \times \frac{5}{6} = \frac{9}{6} - \frac{15}{6} = -1\). 2. On average you lose $1 per game - not a fair game, which would have \(E = 0\). 3. Check: the game pays out \(\frac{1}{6} \times 12 = 2\) dollars on average against a $3 ticket - a $1 house edge. Casinos and insurers price with exactly this calculation.
Variance and standard deviation of a distribution
Spread around \(\mu\) is measured by the variance:
Each squared distance is weighted by its probability, so likely values count heavily and rare extremes count a little. The standard deviation \(\sigma\) is the square root, returning to the variable's own units. Note the contrast with data: a probability distribution weights by \(P(x)\) rather than by counting cases, and divides by nothing - the weights already sum to 1.
Worked example: mean and SD of the green count
The green-count distribution from the previous topic: \(P(0) = \frac{14}{33}\), \(P(1) = \frac{16}{33}\), \(P(2) = \frac{3}{33}\). 1. Mean: \(\mu = 0 \times \frac{14}{33} + 1 \times \frac{16}{33} + 2 \times \frac{3}{33} = \frac{22}{33} = \frac{2}{3}\). 2. Squared distances: \(\frac{4}{9}\), \(\frac{1}{9}\) and \(\frac{16}{9}\). 3. Variance: \(\sigma^2 = \frac{4}{9} \times \frac{14}{33} + \frac{1}{9} \times \frac{16}{33} + \frac{16}{9} \times \frac{3}{33} = \frac{120}{297} = \frac{40}{99} \approx 0.404\). 4. \(\sigma = \sqrt{0.404} \approx 0.64\) greens. Check by the shortcut \(\sigma^2 = E(X^2) - \mu^2\): \(E(X^2) = \frac{16 + 12}{33} = \frac{28}{33}\), and \(\frac{28}{33} - \frac{4}{9} = \frac{40}{99}\).
Rules for means
Two facts make expected values easy to combine. Adding a constant shifts the mean by it, and scaling multiplies it: \(E(a + bX) = a + bE(X)\). And means add without conditions: \(E(X + Y) = E(X) + E(Y)\) for any two random variables, dependent or not. If a cafe's muffin sales \(X\) average 40 and its scone sales \(Y\) average 25, total pastry sales average \(40 + 25 = 65\) no matter how muffins and scones move together.
Rules for variances
Variances transform differently: \(\text{Var}(a + bX) = b^2\text{Var}(X)\). Shifting by \(a\) changes nothing - adding a $5 bonus to every payout moves the centre, not the spread - while scaling by \(b\) multiplies variance by \(b^2\), because distances from the mean scale by \(b\) and are then squared. For sums, variances add when the variables are independent: \(\text{Var}(X + Y) = \text{Var}(X) + \text{Var}(Y)\). Strikingly, variances still add for a difference: \(\text{Var}(X - Y) = \text{Var}(X) + \text{Var}(Y)\). Subtracting independent noise does not cancel; it compounds.
Worked example: transforming a profit
A cafe's daily muffin sales \(X\) have \(\mu = 40\) and \(\sigma = 6\). Each muffin earns $2 and the cafe pays a $10 fixed cost each day, so daily muffin profit is \(Y = 2X - 10\). 1. \(E(Y) = 2(40) - 10 = 70\) dollars. 2. \(\text{Var}(X) = 6^2 = 36\), so \(\text{Var}(Y) = 2^2 \times 36 = 144\). 3. \(\sigma_Y = \sqrt{144} = 12\) dollars. 4. Check: distances from the mean double under the $2-per-muffin scaling, and the constant $10 cost moves every day's profit equally, so the SD doubles exactly - from 6 to 12 - while the mean lands on 70.
Common mistakes to avoid
First, treating the expected value as the most likely value - it is an average, and may be impossible to observe. Second, adding standard deviations: independent variances add, so combine SDs as \(\sqrt{\sigma_1^2 + \sigma_2^2}\), never by simple addition. Third, believing \(\text{Var}(X - Y)\) shrinks: it grows, like any independent sum. Fourth, forgetting the \(b^2\) in \(\text{Var}(a + bX)\) - doubling prices doubles the SD but quadruples the variance. Fifth, computing \(E\) or \(\sigma\) from a table whose probabilities do not total 1. Sixth, adding variances when the variables are not independent.
Key takeaways
The expected value \(E(X) = \sum xP(x)\) is the distribution's balance point and long-run average, even when no single outcome equals it; fair games have \(E = 0\), and the die game's -$1 is a house edge. Variance \(\sigma^2 = \sum (x-\mu)^2P(x)\) and its root \(\sigma\) measure spread, probability-weighted. Means add without conditions: \(E(X + Y) = E(X) + E(Y)\). Linear transforms behave as \(E(a + bX) = a + bE(X)\) and \(\text{Var}(a + bX) = b^2\text{Var}(X)\). Independent variances add - even for differences - and standard deviations never do.
Flashcards
Flashcard
Expected value: the probability-weighted average
Test your understanding
Question 1 of 5
X takes values 0, 1 and 2 with probabilities 0.1, 0.4 and 0.5. What is E(X)?
Question 2 of 5
You pay $2 to roll a die and win $10 on a six (net +8), nothing otherwise (net -2). What is the expected net gain per game, to the nearest cent?
Question 3 of 5
Var(X) = 4. What is Var(3X + 5)?
Question 4 of 5
X and Y are independent with variances 9 and 16. What is Var(X + Y)?
Question 5 of 5
A bakery sells X loaves daily with E(X) = 30, and daily loaf profit is Y = 3X - 15 dollars. What is E(Y)?
