Five Randomness Experiments You Can Run in a Classroom Today
The best way to learn probability is to watch it happen. Textbook problems can teach the formulas, but the intuitions that make probability genuinely useful — the gut-level understanding of why streaks happen, why rare events aren't as rare as they feel, why averages stabilize over time — develop through direct experience with random processes. The following five experiments require no special equipment, work for students from late elementary through college, and teach core concepts that show up in statistics, science, economics, and everyday reasoning.
The Streak Experiment. Before class, privately flip a coin 100 times and record the exact sequence of heads and tails. Then ask students to write down what they think a random sequence of 100 coin flips would look like — no actual flipping, just their best guess at a plausible random sequence. Collect all the sequences, including your real one, and display them without labels. Ask the class to identify which one is the real sequence. In almost every run of this experiment, students can pick out the real sequence, and the reason is instructive: the real sequence contains longer streaks (five, six, seven identical outcomes in a row) than any of the fabricated ones. Humans instinctively avoid long streaks when simulating randomness because streaks don't feel random, but genuine randomness produces them regularly. The experiment makes the gap between perceived and actual randomness visible in a way that no lecture can.
The Law of Large Numbers, Live. Give each student a coin (physical or virtual) and ask them to flip it 10 times, recording the number of heads. On the board, plot each student's result. The individual results will vary widely — some students will get 3 heads, others 7, a few might get 2 or 8. Now have each student flip 50 more times and record the cumulative proportion of heads across all 60 flips. Plot the new proportions. They'll be clustered much more tightly around 50%. If time allows, continue to 100 total flips. The convergence toward the expected value becomes visible in real time, and the key lesson — that averages stabilize with sample size but individual outcomes remain unpredictable — is something students can see in their own data rather than taking on faith.
The Birthday Problem. Ask students to guess how many people you'd need in a room before there's a better-than-even chance that two of them share a birthday. Most people guess something close to 183, reasoning that you need about half of 365. The actual answer is 23, and the surprise on students' faces when they hear this is the best hook in probability education. The reason the number is so low is that the question isn't about matching a specific birthday — it's about matching any pair, and the number of possible pairs grows much faster than the number of people. With 23 people, there are 253 possible pairs, each of which is an independent opportunity for a match. If your class has more than 23 students, test it live. There's a better-than-even chance you'll find a shared birthday in the room, and when you do, the abstract calculation becomes a concrete, memorable event. For smaller classes, simulate it: use a random number generator set to 1–365 to generate 23 "birthdays" and check for duplicates. Repeat ten or twenty times and count how often a duplicate appears.
The Monty Hall Problem. This one requires a little setup but pays enormous dividends in teaching conditional probability. Place a prize behind one of three doors (cups, folders, or envelopes work). A student picks a door. You, the host, open one of the remaining doors that you know is empty. The student then chooses whether to stick with their original pick or switch to the other unopened door. The correct strategy is to always switch — switching wins 2/3 of the time, while staying wins only 1/3 — but this result is so counterintuitive that even professional mathematicians have argued about it. The way to make it click is to play the game many times. Have students pair up and play 30 rounds each, 15 using the "always stay" strategy and 15 using "always switch," and record the win rates. When the class data is pooled, the 2/3 vs. 1/3 split emerges clearly, and students who were skeptical of the math are confronted with their own empirical evidence. The lesson that the correct answer can feel wrong, and that running the experiment is the way to resolve the conflict between intuition and logic, is one of the most valuable things a probability course can teach.
Sampling Bias, Demonstrated. Fill a bag or opaque container with colored marbles (or folded slips of paper with colors written on them) in a known ratio — say, 60% blue and 40% red. Don't tell students the ratio. Have each student draw a sample of 5 marbles (replacing each marble before the next draw), estimate the percentage of blue marbles, and write down their guess. Plot the guesses. They'll be scattered widely, with some students guessing 20% blue and others guessing 100%. Now have each student draw a sample of 30 (still with replacement) and re-estimate. The guesses tighten dramatically. Finally, pool all the class data into one large sample and calculate the overall proportion. It will be very close to 60%. The experiment teaches three things simultaneously: small samples are unreliable, large samples are more accurate, and pooling data improves estimates. These are the foundational ideas behind polling, clinical trials, and quality control, and they land far more effectively when students discover them from their own data than when they read them in a textbook.
Each of these experiments takes between fifteen and thirty minutes, uses tools that cost nothing (a few coins, some paper, and a screen showing a dice roller or number generator), and produces genuine moments of surprise that help students remember the underlying principle long after the class is over. The best math teaching doesn't tell students what to expect — it sets up conditions where they can discover it themselves, and randomness, by its nature, is full of discoveries.