Empirical vs theoretical probability
By Jude Wallis · Published
Theoretical probability is found by reasoning: with equally likely outcomes, it is the number of outcomes in the event divided by the total number of outcomes. Empirical probability is found from data: the number of times the event occurred divided by the number of trials.
AP Statistics: Unit 2 (topics 2.3 Estimating Probabilities Using Simulation, 2.4 Introduction to Probability). In the Fall 2026 AP Statistics course, estimating a probability from the relative frequency in many trials is topic 2.3, Estimating Probabilities Using Simulation, which also states the law of large numbers. Calculating a probability from equally likely outcomes is topic 2.4, Introduction to Probability. Empirical and experimental are common textbook names for the first approach. Both topics are in Unit 2.
Empirical vs theoretical probability: the short answer
The two are different routes to the probability of the same event.
- Theoretical probability uses a model of the random process. No trials are run. For a fair die, the model says six outcomes are equally likely, so .
- Empirical probability uses results. You run the process many times, or simulate it, and compute the relative frequency of the event. If a 4 comes up 21 times in 120 rolls, the empirical probability is .
A theoretical probability is a fixed number. An empirical probability is an estimate, and it changes from one set of trials to the next. With more trials, the estimate tends to settle closer to the true probability.
What theoretical probability is
When all outcomes in the sample space are equally likely, the theoretical probability of an event is
The work is counting. For two fair dice there are equally likely ordered outcomes, and 6 of them have a sum of 7, so .
The method depends on the model being right. Counting outcomes gives a valid probability only when the outcomes really are equally likely. The eleven possible sums of two dice, 2 through 12, are not equally likely, so the answer for a sum of 7 is wrong. Probability rules such as the complement rule and the multiplication rule also produce theoretical probabilities, since they reason from a model without collecting data.
What empirical probability is
Empirical probability is the relative frequency of an event in a set of trials:
The trials can be real, such as 500 flips of a bottle cap, or simulated with random digits, dice, or software. It is also called experimental probability.
This ties to how probability is defined in the first place: the probability of an event is its long-run relative frequency. The law of large numbers says that for independent trials, the relative frequency gets closer and closer to a single value, the true probability, as the number of trials increases. So an empirical probability from many trials is a good estimate, and one from a handful of trials can be far off. Designing the trials is covered in how to simulate a random process.
The differences side by side
| Feature | Theoretical probability | Empirical probability |
|---|---|---|
| Based on | A model and counting or probability rules | Observed or simulated trials |
| Formula | Outcomes in the event over total outcomes, when equally likely | Times the event occurred over number of trials |
| Needs data | No | Yes |
| Changes if you repeat the work | No | Yes, each set of trials gives a new value |
| Main requirement | The model must be correct | Enough trials, run under the same conditions |
| Also called | Classical probability | Experimental probability, relative frequency |
| Sum of 7 on two dice | in one run of 200 rolls |
Neither column is the "real" one in every case. Where a trustworthy model exists, the theoretical value is the true probability and the empirical value estimates it. Where no model exists, the empirical value is the best available.
When to use which
Use theoretical probability when the outcomes can be counted and are equally likely, or when the problem gives enough probabilities to apply the rules: fair coins, fair dice, a shuffled deck, a random draw from a known list.
Use empirical probability in two situations. The first is when no model is available. A bottle cap, a thumbtack, or a basketball player's free throw has no set of equally likely outcomes to count, so the only way to learn the probability is to observe many trials. The second is when a model exists but the calculation is hard. Simulating the process many times and taking the relative frequency gives an estimate without the algebra.
The two also check each other. If you roll a die 600 times and a 6 appears 190 times, the empirical probability of about 0.317 is far from the theoretical , and that gap is evidence against the fair-die model.
The classic mix-up and how to avoid it
The first mistake is expecting a small number of trials to match the theoretical value. Ten flips of a fair coin produce exactly 5 heads only about 24.6% of the time. An empirical result of 7 heads in 10 does not show the coin is unfair, and it does not change the theoretical probability of 0.5.
The second mistake is the reverse: treating an empirical probability as exact. A result of is an estimate of the probability of a sum of 7, and a new run of 200 rolls would give a different number. Report it as an estimate and say how many trials it rests on.
The third mistake is believing short runs must balance out. The law of large numbers describes the relative frequency over a long run. It does not make tails more likely after a streak of heads, because independent trials have no memory. The difference between that law and the central limit theorem is laid out in law of large numbers vs central limit theorem.
A sum of 7 on two dice, both ways
Find the theoretical probability that two fair dice show a sum of 7. Then compare it with two sets of rolls: 38 sums of 7 in 200 rolls, and 329 sums of 7 in 2,000 rolls.
Count the sample space: each die has 6 faces, so there are equally likely ordered outcomes.
Count the outcomes with a sum of 7: (1,6), (2,5), (3,4), (4,3), (5,2), (6,1). That is 6 outcomes.
Theoretical probability: .
Empirical probability from 200 rolls: . The gap from the theoretical value is .
Empirical probability from 2,000 rolls: . The gap is .
Judge the gaps against chance variation. The standard deviation of the relative frequency is , which is about for and about for . Both gaps are smaller than one standard deviation, so both results are consistent with fair dice.
The theoretical probability is . The empirical probabilities are from 200 rolls and from 2,000 rolls. Both are estimates of the same value, and the estimate from the larger number of rolls landed closer, which is the usual pattern.
A probability with no model to count
A student flips a plastic bottle cap 500 times and it lands with the open side up 185 times. Estimate the probability that the cap lands open side up, and explain why a theoretical calculation is not available.
Identify the event and the trials: the event is 'lands open side up', and there are 500 trials.
Compute the relative frequency: .
Consider a theoretical approach. The cap has two resting positions, but its shape and weight are not symmetric, so there is no reason to treat the two positions as equally likely. Counting outcomes would give , which rests on an assumption the cap does not satisfy.
Compare the data with that assumption. If the true probability were 0.5, 500 flips would give about 250 open-side-up results. The observed 185 is far below that, which supports rejecting the equally likely model.
State the estimate with its basis: about 0.37, from 500 flips.
The empirical probability is . No theoretical value can be counted, because the two ways the cap can land are not equally likely. The estimate of 0.37 would change slightly with a new set of flips.
Frequently asked questions
Why is my empirical probability different from the theoretical probability?
Because an empirical probability comes from a limited number of random trials, and random results vary. A gap is expected, and it tends to shrink as the number of trials grows. A large gap that persists over many trials suggests the theoretical model is wrong.
Which is more accurate, empirical or theoretical probability?
When the model is correct, the theoretical probability is exact and the empirical one is an estimate of it. When no reliable model exists, as with a thumbtack or a bottle cap, the empirical probability from many trials is the only sound choice.
Is a probability from a simulation empirical or theoretical?
Empirical. A simulation runs many trials of a modeled process and reports a relative frequency, so its result is an estimate that varies from run to run, even though a model was used to generate the trials.