Can experimental probability ever equal theoretical probability exactly?

Yes. If you flip a fair coin 50 times and get exactly 25 heads, the experimental probability is 25/50 = 0.5, matching the theoretical value. It just is not guaranteed, and it becomes less likely to hit exactly as the number of trials grows, even though the two values get closer overall.

How many trials are enough?

There is no single magic number, but more is always better, and hundreds are far more convincing than dozens. A useful gut check: with n trials, one extra occurrence moves the experimental probability by 1/n. If that shift is large compared to the gap you are studying, you need more trials.

If I get five tails in a row, is heads more likely next?

No. Each flip is independent, so the probability of heads on the very next flip is still 1/2. The coin has no memory of past flips. What evens out over time is the long-run fraction, because later trials add to a growing denominator — not because the coin owes you anything.

Which probability should I use to make a prediction?

If the model is trustworthy — a fair cube, equal spinner sections, a shuffled deck — use the theoretical probability. If the object is irregular or the situation has no obvious equally likely outcomes, use the experimental probability from as many trials as you have.

M7MATH-10.2

Experimental vs Theoretical Probability

Learn to compute experimental probability from trial data, compare it with theoretical probability, and use probability to predict how often an event happens.

What you'll do in this lesson

A voice-first session with the Crimsora tutor on Experimental vs Theoretical Probability, then targeted practice and FRQs — with the tutor adapting to where you get stuck.

Start Experimental vs Theoretical Probability free

What this lesson covers

Flip a fair coin 10 times and you might get 7 heads. Does that mean the coin is broken? No — it means real data wobbles around what the math predicts. In this lesson you will learn two different ways to put a number on chance: theoretical probability, which comes from thinking about the outcomes in a model, and experimental probability, which comes from actually running trials and counting what happened.

You will practice computing both, comparing them, explaining why they rarely match exactly, and describing what happens to the gap between them as you collect more and more trials. Then you will flip the idea around and use a probability to predict — if the probability of a red marble is

\frac{2}{5}

, about how many reds should you expect in 200 draws? This prediction skill shows up constantly in science labs, sports statistics, and quality checks in factories.

Two Ways to Measure Chance

Theoretical probability is what a model predicts. You list the possible outcomes, decide which ones count as your event, and compute

P(\text{event}) = \frac{\text{number of favorable outcomes}}{\text{total number of equally likely outcomes}}

Rolling a 5 on a fair number cube has theoretical probability

\frac{1}{6}

because there is one 5 among six equally likely faces. You never have to roll anything to find this number — you reason about the cube.

Experimental probability (also called relative frequency) is what actually happened. You run trials, count, and compute

P(\text{event}) = \frac{\text{number of times the event occurred}}{\text{total number of trials}}

If you roll the cube 30 times and get a 5 exactly 7 times, the experimental probability of a 5 is

\frac{7}{30}

, about

0.233

	Theoretical	Experimental
Comes from	a model of the situation	data from real trials
Denominator	total possible outcomes	total number of trials
Needs equally likely outcomes?	yes, for the counting formula	no
Changes when you collect more data?	no	yes

Notice the last row. Theoretical probability of heads is

\frac{1}{2}

whether you flip once or a million times. Experimental probability is recalculated every time you add trials, so it is a moving number.

One more important point: experimental probability works even when you cannot list equally likely outcomes. Nobody can compute from first principles the probability that a thumbtack lands point-up, or that a particular basketball player makes a free throw. For those, data is the only tool you have.

Why the Two Numbers Differ

Students often assume that if theory says

\frac{1}{2}

, then 20 flips must give exactly 10 heads. Real data almost never behaves that politely. Getting 12 heads out of 20 gives an experimental probability of

\frac{12}{20} = 0.6

, which is not equal to

0.5

— and nothing is wrong.

The reason is random variation. Each individual trial is unpredictable. In a short run, a few lucky outcomes can pull the fraction noticeably away from the theoretical value because the denominator is small. With 10 flips, one extra head changes the experimental probability by

0.1

. With 1000 flips, one extra head changes it by

0.001

. The same amount of luck matters far less when it is spread over many trials.

This leads to the big idea, sometimes called the law of large numbers: as the number of trials increases, the experimental probability tends to get closer to the theoretical probability. The gap narrows — not because later trials "correct" earlier ones, but because a bigger denominator swallows the wobble.

Watch this common misconception. After five heads in a row, many students say a tail is "due." The coin has no memory. The probability of a tail on the next flip is still

\frac{1}{2}

. What the law of large numbers promises is about the long-run fraction, not about the next single outcome.

A second misconception: a large gap always means something is unfair. With only 8 trials, a gap is expected. But if you spin a spinner 500 times and red comes up

70\%

of the time when the model says

25\%

, that gap is too big to explain by luck — that is real evidence the model is wrong or the spinner is weighted.

Using Probability to Predict

Probability runs in both directions. Given a probability and a number of trials, you can predict how often an event should happen:

\text{predicted count} = P(\text{event}) \times \text{number of trials}

If the probability of drawing a green marble is

\frac{3}{10}

and you draw 250 times with replacement, you predict

\frac{3}{10} \times 250 = 75

greens. This is an expected count, not a guarantee. Getting 71 or 82 would be completely ordinary; getting 5 would be shocking.

You can use either kind of probability as the input. A theoretical probability gives a prediction based on the model. An experimental probability gives a prediction based on past data — which is what a coach does when a player who has made 18 of 25 free throws is expected to make about

\frac{18}{25} \times 40 = 28.8

, roughly 29, of her next 40 attempts.

Two places students slip up here. First, they add instead of multiply, or they divide the trials by the probability. Anchor yourself with a check: if the probability is about half, the prediction should be about half the trials. Second, they round too aggressively or forget that a predicted count can come out as a decimal. Saying "about 29 free throws" is fine; saying "exactly 28.8 free throws" is not, since you cannot make part of a shot.

You can also work backward. If you expect 12 sixes and the probability of a six is

\frac{1}{6}

, then the number of rolls satisfies

\frac{1}{6}n = 12

, so

n = 72

Setting Up and Reading a Trial Table

Most classroom problems hand you a frequency table from an experiment. Reading it carefully is half the work.

Suppose a bag holds colored chips. A student draws one chip, records the color, returns it, and repeats 60 times:

Color	Frequency	Experimental probability
Red	27	$\frac{27}{60} = 0.45$
Blue	18	$\frac{18}{60} = 0.30$
Yellow	15	$\frac{15}{60} = 0.25$

Three habits keep this accurate. First, find the total by adding all frequencies — here

27 + 18 + 15 = 60

— instead of assuming a number from the problem. Second, check that your experimental probabilities add to

1

0.45 + 0.30 + 0.25 = 1

. If they do not, you used the wrong denominator somewhere. Third, keep the fractions in a comparable form. To compare

\frac{27}{60}

with a theoretical

\frac{1}{3}

, convert both to decimals (

0.45

versus about

0.333

) or to a common denominator. Comparing raw counts to fractions is the single most common error in these problems.

If the model says the three colors are equally likely, theoretical probability for each is

\frac{1}{3} \approx 0.333

. Red is running high and yellow low. With 60 trials, gaps of this size are plausible but worth investigating — if 600 trials showed the same pattern, you would conclude the chips are not in equal numbers and would trust the experimental values over the equal-likelihood model.

Deciding Which Probability to Trust

When theory and data disagree, which one should guide you? It depends on how much data you have and whether the model is believable.

If you have a strong reason to believe the model — a factory-made number cube, a carefully divided spinner, a shuffled deck — trust the theoretical probability for predictions and treat small gaps in data as ordinary variation.

If the model is a guess, or if the object might be irregular (a bent cap, a weighted die, a thumbtack), the data is your best information. More trials means a more trustworthy experimental probability.

A useful way to think about it: theoretical probability tells you what should happen; experimental probability tells you what did happen. Comparing them is how scientists test whether a model describes reality. If you flip a coin 400 times and get 203 heads, the experimental probability is

\frac{203}{400} = 0.5075

, extremely close to

0.5

— strong support for "fair coin." If you get 310 heads,

\frac{310}{400} = 0.775

, that is far too extreme for luck across 400 trials, and you should suspect the coin.

When you write up a comparison, a complete answer does three things: states both probabilities as numbers in the same form, says which is larger and by how much, and explains the difference in terms of the number of trials. Simply writing "they are different" leaves out the reasoning that matters.

Key terms

Theoretical probability.: The probability predicted by a model, computed as favorable outcomes divided by total equally likely outcomes. It does not require doing any trials.
Experimental probability.: The probability estimated from data, computed as the number of times the event occurred divided by the total number of trials. Also called relative frequency.
Trial.: One repetition of a chance process, such as a single coin flip, one spin of a spinner, or one draw from a bag.
Relative frequency.: A count divided by the total number of trials, expressed as a fraction, decimal, or percent — the same value as experimental probability.
Law of large numbers.: The principle that as the number of trials increases, the experimental probability tends to get closer to the theoretical probability.
Random variation.: The natural, unpredictable differences between what a model predicts and what actually happens in a limited number of trials.
Expected count.: The predicted number of occurrences of an event, found by multiplying the probability by the number of trials. It is an estimate, not a guarantee.
Equally likely outcomes.: Outcomes that each have the same chance of occurring, which is the condition needed for the counting formula for theoretical probability.

Worked example

A spinner has 4 equal sections labeled A, B, C, and D. Maya spins it 80 times and records: A 26 times, B 14 times, C 19 times, D 21 times. (a) Find the theoretical probability of landing on A. (b) Find the experimental probability of landing on A. (c) Compare them and explain the difference. (d) Predict how many times A would come up in 400 spins.

(a) Theoretical probability. The four sections are equal, so the four outcomes are equally likely. There is 1 favorable outcome (A) out of 4 total:

P(A) = \frac{1}{4} = 0.25

(b) Experimental probability. First confirm the total number of trials by adding the frequencies:

26 + 14 + 19 + 21 = 80

. That matches the 80 spins stated. Now divide the count for A by the total:

P(A) = \frac{26}{80} = \frac{13}{40} = 0.325

(c) Compare. Put both in the same form: theoretical

0.25

, experimental

0.325

. The experimental probability is higher, by

0.325 - 0.25 = 0.075

, or 7.5 percentage points. The difference is caused by random variation. In 80 spins the model predicts

0.25 \times 80 = 20

landings on A, and Maya got 26 — six more than predicted. Six extra out of 80 trials is an ordinary amount of luck, not evidence the spinner is unfair. If she kept spinning to 800 or 8000 times, the experimental probability would be expected to drift closer to

0.25

.

(d) Predict for 400 spins. Use the theoretical probability, since the spinner sections are known to be equal:

0.25 \times 400 = 100

Expect A about 100 times. Note this is an expected count — results like 92 or 107 would be perfectly normal. If instead you based the prediction on Maya's data, you would get

0.325 \times 400 = 130

, but the equal-section model is more trustworthy here than 80 spins of data.

Practice questions

A number cube is rolled 150 times and lands on 4 a total of 31 times. Which statement correctly compares the experimental and theoretical probabilities of rolling a 4?

The experimental probability is about $0.207$ , slightly higher than the theoretical $0.167$ .
The experimental probability is $\frac{1}{6}$ , exactly equal to the theoretical probability.
The experimental probability is $31$ , and the theoretical probability is $25$ .
The experimental probability is about $0.207$ , which proves the cube is unfair.

Answer: The experimental probability is about $0.207$ , slightly higher than the theoretical $0.167$ .

Experimental probability is

\frac{31}{150} \approx 0.207

. Theoretical probability is

\frac{1}{6} \approx 0.167

. So the experimental value is a bit higher. The third choice confuses counts with probabilities — probabilities are always between 0 and 1. The last choice overreaches: the model predicts

\frac{1}{6} \times 150 = 25

fours, and getting 31 is only 6 above the prediction, an ordinary amount of random variation across 150 rolls, not proof of unfairness.

A bag contains 5 red, 3 blue, and 2 green marbles. A marble is drawn and replaced 400 times. Predict how many blue marbles you expect, and explain why the actual result will probably not equal your prediction exactly.

Answer: About 120 blue marbles. The prediction is $P(\text{blue}) \times 400 = \frac{3}{10} \times 400 = 120$ . The actual count will likely differ because each draw is random.

There are

5 + 3 + 2 = 10

marbles, so

P(\text{blue}) = \frac{3}{10} = 0.3

. Multiply by the number of trials:

0.3 \times 400 = 120

. This is an expected count, an average of what would happen over many sets of 400 draws — not a guarantee. Random variation means a real run might give 112 or 129 blue marbles. What the law of large numbers says is that the experimental probability

\frac{\text{blues}}{400}

should land close to

0.3

, and closer still if you did 4000 draws.

Liam flips a bottle cap 20 times and it lands upright 6 times. His theoretical model assumed upright and not-upright were equally likely. He says the gap proves his model is wrong. Sofia flips the same cap 500 times and gets 148 upright. What should they conclude, and why is Sofia's evidence stronger?

Answer: Liam's 20 flips give $\frac{6}{20} = 0.30$ and Sofia's 500 flips give $\frac{148}{500} = 0.296$ . Both are far from $0.5$ , but Sofia's result is stronger evidence because a large number of trials makes random variation a much less believable explanation. They should reject the equally-likely model and use about $0.30$ as the probability.

With only 20 flips, a result of 6 upright could plausibly happen by luck even if the true probability were

0.5

, since one flip shifts the fraction by

0.05

. With 500 flips, one flip shifts the fraction by only

0.002

, so a value of

0.296

cannot reasonably be luck if the truth were

0.5

. This also shows why a bottle cap needs experimental probability in the first place: its two positions are not physically equally likely, so the counting formula for theoretical probability does not apply.

FAQ

Can experimental probability ever equal theoretical probability exactly?: Yes. If you flip a fair coin 50 times and get exactly 25 heads, the experimental probability is $\frac{25}{50} = 0.5$ , matching the theoretical value. It just is not guaranteed, and it becomes less likely to hit exactly as the number of trials grows, even though the two values get closer overall.
How many trials are enough?: There is no single magic number, but more is always better, and hundreds are far more convincing than dozens. A useful gut check: with $n$ trials, one extra occurrence moves the experimental probability by $\frac{1}{n}$ . If that shift is large compared to the gap you are studying, you need more trials.
If I get five tails in a row, is heads more likely next?: No. Each flip is independent, so the probability of heads on the very next flip is still $\frac{1}{2}$ . The coin has no memory of past flips. What evens out over time is the long-run fraction, because later trials add to a growing denominator — not because the coin owes you anything.
Which probability should I use to make a prediction?: If the model is trustworthy — a fair cube, equal spinner sections, a shuffled deck — use the theoretical probability. If the object is irregular or the situation has no obvious equally likely outcomes, use the experimental probability from as many trials as you have.

Learn this with a teacher, not a page

The Crimsora tutor teaches Experimental vs Theoretical Probability live — explaining on a whiteboard, asking you questions, and adapting to where you get stuck.

Try a live session free