Expected Value as a Probability-Weighted Average

Why Weight Outcomes by Their Probabilities?

A random variable can take several possible values, but those values may not be equally likely. If we average the distinct values without considering how often each occurs, a rare outcome counts just as much as a common one. Expected value accounts for this difference: it combines each possible value with the probability of seeing it.

For example, suppose a waiting time can be 0, 5, 10, or 15 minutes. If most waits are short and a 15-minute wait is unusual, treating all four times equally would not describe the typical result over many appointments. The expected value is designed to summarize the distribution by weighting outcomes according to their probabilities.

Definition: Let \(X\) be a discrete random variable with a finite set of possible values \(x_1,\ldots,x_k\), and let \(P(X=x_i)=p_i\). Its expected value (also called its population mean) is

E(X)=\sum_{i=1}^{k} x_i p_i.

Each probability satisfies \(p_i\geq 0\), and the probabilities of all possible values sum to 1. Thus, expected value is the probability-weighted average of the possible values.

The notation \(\sum_{i=1}^{k}x_i p_i\) means to calculate \(x_i p_i\) for each possible value and then add all the products. The values \(x_i\) must be the distinct possible values of \(X\), and each \(p_i\) must be the probability of the corresponding value. If the probabilities do not sum to 1, they do not form a valid probability distribution, so the calculation cannot be used as an expected value.

For a random variable with infinitely many possible values, the corresponding definition uses an infinite sum when that sum is well-defined. This tutorial focuses on finite discrete distributions, for which the calculation is a finite sum.

Calculating an Expected Value

The calculation is systematic. First list every possible value of \(X\) and its probability. Check that the probabilities are nonnegative and sum to 1. Then multiply each value by its probability and add the products. The units of \(E(X)\) are the same as the units of \(X\), because the probabilities are unitless.

List the outcomes and probabilities.
Include all possible values of the random variable, with the probability attached to each value.

Check the distribution.
Verify that each probability is between 0 and 1 and that the probabilities add to 1.

Weight and add.
Multiply each outcome by its probability, then sum those products and report the result with the variable’s units.

Worked Example: Expected Waiting Time

Suppose, hypothetically, that the waiting time \(X\) for an appointment has this distribution: 0 minutes with probability 0.20, 5 minutes with probability 0.40, 10 minutes with probability 0.30, and 15 minutes with probability 0.10. Find \(E(X)\) and compare it with the unweighted average of the listed times.

Check the probabilities. They are all between 0 and 1, and \(0.20+0.40+0.30+0.10=1.00\), so they form a valid distribution.

Calculate the expected value. Multiply each waiting time by its probability and add:

E(X)=(0)(0.20)+(5)(0.40)+(10)(0.30)+(15)(0.10) =0+2+3+1.5=6.5\text{ minutes}.

The expected waiting time is 6.5 minutes. By contrast, the unweighted average of the four distinct times is \((0+5+10+15)/4=7.5\) minutes. That calculation gives each distinct time equal weight, even though the distribution assigns different probabilities to them. It is not the expected waiting time.

Expected Value as a Long-Run Average

Expected value is a feature of a probability distribution, not a prediction that the next observation will equal \(E(X)\). Its long-run interpretation comes from imagining that the same random process is repeated many times under stable conditions. If an outcome is more probable, it tends to occur more often; in a long sequence, the proportion of times each value occurs tends to be close to its probability. The average of the observed values then approaches the probability-weighted average.

This interpretation assumes repeated trials are independent and have the same probability distribution, with a finite expected value. Independence here means that the result of one trial does not change the probabilities for another trial. Under those conditions, the law of large numbers says that the observed sample average approaches \(E(X)\) as the number of trials grows. It does not say that a short sequence must average exactly to \(E(X)\), or that every individual result will be close to it.

Why the Long-Run Average Has This Form

The connection can be seen by writing the average from repeated observations in terms of how often each value occurred. Suppose \(X\) has possible values \(x_1,\ldots,x_k\). In \(N\) independent repetitions with the same distribution, let \(n_i\) be the number of repetitions in which the result was \(x_i\). The total of all observed values is \(x_1n_1+\cdots+x_kn_k\), so the sample average is:

\bar{x} =\frac{x_1n_1+\cdots+x_kn_k}{N} =\sum_{i=1}^{k}x_i\frac{n_i}{N}.

Here, \(n_i/N\) is the observed relative frequency of outcome \(x_i\). As the number of repetitions increases, the law of large numbers says that these relative frequencies approach their probabilities \(p_i\). Since there are only finitely many outcomes, replacing each relative frequency by its limiting probability gives:

\bar{x}\ \longrightarrow\ \sum_{i=1}^{k}x_i p_i=E(X).

This is why the probability-weighted average, rather than the unweighted average of the listed outcomes, describes the long-run average result.

Worked Example: A Distribution With Unequally Likely Scores

Suppose a hypothetical practice task assigns a score \(X\), measured in points, with probabilities \(P(X=-2)=0.25\), \(P(X=1)=0.50\), and \(P(X=4)=0.25\). Find the expected score and assess whether the unweighted average gives the same answer.

Check the probabilities. Each is between 0 and 1, and \(0.25+0.50+0.25=1.00\).

Calculate the probability-weighted average.

E(X)=(-2)(0.25)+(1)(0.50)+(4)(0.25) =-0.50+0.50+1.00=1.00\text{ point}.

The unweighted average of the three possible scores is \((-2+1+4)/3=1\) point, which happens to match the expected value here. The match is coincidental: the score of 1 has twice the probability of either other score, and those probabilities happen to balance the values in this distribution. In general, the unweighted average is not justified unless the possible values are equally likely (or the particular probabilities and values happen to produce the same result).

Expected Value Need Not Be a Possible Outcome

An expected value is an average, so it need not be one of the values the random variable can actually take. For example, a count of events must be a whole number on any particular occasion, but its expected value can be a fraction of an event per occasion. That fractional value describes an average across occasions; it does not describe a partial event.

The expected value must lie between the smallest and largest possible values when the distribution has finite support. To see why, let \(x_{\min}\) and \(x_{\max}\) be the smallest and largest possible values. Each value satisfies \(x_{\min}\leq x_i\leq x_{\max}\). Multiplying each inequality by the nonnegative \(p_i\) preserves its direction. Adding across all outcomes gives:

x_{\min}\sum_{i=1}^{k}p_i \leq \sum_{i=1}^{k}x_i p_i \leq x_{\max}\sum_{i=1}^{k}p_i.

Because the probabilities sum to 1, this becomes \(x_{\min}\leq E(X)\leq x_{\max}\). This bound is a useful check on arithmetic, but it does not require the expected value to equal one of the possible values.

Worked Example: An Expected Count Below One

Suppose, hypothetically, that \(X\) is the number of appointment cancellations in a group during one morning. Its distribution is \(P(X=0)=0.60\), \(P(X=1)=0.30\), and \(P(X=2)=0.10\). Find the expected number of cancellations and interpret it.

Check the probabilities. All are nonnegative and no greater than 1, and \(0.60+0.30+0.10=1.00\).

Calculate the expected count.

E(X)=(0)(0.60)+(1)(0.30)+(2)(0.10) =0+0.30+0.20=0.50\text{ cancellations per morning}.

The value 0.50 is not a possible count for one morning: the count must be 0, 1, or 2. It means that across many mornings with this same distribution, the average number of cancellations per morning would approach 0.50. The unweighted average of the possible counts is \((0+1+2)/3=1\), which is not the probability-weighted average because the three counts are not equally likely.

Common Mistakes and Why the Distinction Matters

Expected value is useful whenever a probability distribution needs a single measure of its average location. In the long run, it provides a reference value for the results produced by a repeated process. In a short run, outcomes can differ substantially from it, especially if the distribution puts nontrivial probability on values far from the mean. Expected value summarizes the distribution; it does not describe every trial or guarantee what will happen next.

Averaging the listed values without probabilities. This treats each distinct value as equally likely. Instead, multiply each value by its own probability and add.
Forgetting a probability. Each possible value must be paired with its corresponding probability. Check that the probabilities cover the full distribution and sum to 1.
Interpreting the expected value as a guaranteed result. The next observation may be above or below it, and the expected value may not be a possible outcome at all.
Dropping the units. If \(X\) measures minutes, \(E(X)\) is in minutes; if \(X\) counts events, the expected value is an average count per trial or occasion.

A probability-weighted average gives more influence to outcomes that are more likely. The ordinary average of the distinct possible values gives them equal influence, regardless of probability. The two calculations answer different questions; for a random variable, the probability-weighted calculation is the expected value.

Key takeaway: For a finite discrete random variable, calculate \(E(X)\) by multiplying every possible value by its probability and summing the products. Under repeated independent trials from the same distribution, this value is the long-run average, not a promise about any one result.

Check Your Understanding

For each question, show enough work to explain your reasoning.

A random variable takes values 2, 6, and 10 with probabilities 0.50, 0.30, and 0.20. Calculate its expected value and check that the probabilities form a valid distribution.
For the distribution in question 1, calculate the unweighted average of the three distinct values. Why is it different from the expected value?
A count \(X\) can be 0 with probability 0.70 or 1 with probability 0.30. Find \(E(X)\). Is the expected value a possible value of \(X\), and what does it mean over many repetitions?
State the conditions behind interpreting expected value as a long-run average.
Explain why the expected value of a finite discrete random variable must be between its smallest and largest possible values.