The Shape of Randomness

1.1 When Randomness Isn't Random

The Experiment

Imagine you sit down at a table with a coin. You flip it. Heads. You flip it again. Tails. Again. Heads. Heads. Tails.

Each flip is a tiny mystery — you genuinely cannot predict whether it'll land heads or tails. No amount of concentration, analysis, or wishful thinking will tell you what the next flip will be.

Now imagine you keep going. You flip that coin 1,000 times and record every result. After all 1,000 flips, you count up the heads and make a simple bar chart: one bar for heads, one bar for tails.

Here's the strange part: if you repeated this entire 1,000-flip experiment tomorrow, your bar chart would look almost identical. And the day after that. And the day after that.

How is that possible? Each individual flip is completely unpredictable. So why does the overall picture always come out the same?

Pause and think: If each flip is truly random, why would 1,000 flips give a predictable result? What's your intuition — why does randomness seem to "settle down" when you have enough of it?

That tension — between the chaos of each individual flip and the order of the whole — is the central mystery of this entire course. By the end, you'll have a precise language for describing that order. But right now, let's just watch it happen.

Your First Prediction

Before we explore further, commit to a guess.

Prediction: You flip a fair coin 100 times. What proportion of flips do you think will be heads? Pick one:

(a) Exactly 50%
(b) Close to 50%, but probably not exactly
(c) It could be anything — it's random
(d) Somewhere between 40% and 60%, but I'm not sure why

Lock in your answer. We'll come back to it.

Watch Randomness Unfold

Let's start with something small enough to see clearly.

Try this:

1. Flip 10 times. Look at the proportion of heads. Is it close to 50%? Probably not. You might see 30% or 70% — it's all over the place.

2. Flip up to 50 total. Watch the proportion. It's still bouncing around, but maybe the bounces are getting smaller?

3. Flip up to 200 total. Now look at the line graph of the proportion over time. What's happening?

4. Flip up to 1,000 total. Where is the proportion settling?

What did you notice? The proportion of heads probably started wild — maybe 0.3 or 0.7 after just 10 flips — but it gradually settled closer and closer to 0.5. The more flips, the less it wanders.

Now here's the key move. Reset and do it again. The exact sequence of H's and T's will be completely different. But watch that proportion line. It does the same thing: starts wild, then converges.

The individual flips are different every time. The pattern is the same every time.

From One to Many: Where Order Comes From

Let's slow down and understand why this happens. It's not magic — it's arithmetic.

Consider a tiny example. You've flipped a coin 10 times and gotten 7 heads. That's a proportion of $\frac{7}{10} = 0.70$ — pretty far from 0.50.

Now you flip 10 more times. In these new flips, you happen to get 4 heads (pretty typical — not exactly 5, but close). Your new totals: 11 heads out of 20 flips.

$$\text{Proportion of heads} = \frac{11}{20} = 0.55$$

Look what happened. Even though the second batch wasn't "perfectly balanced," the overall proportion moved from 0.70 down to 0.55. It got pulled closer to 0.50.

Why? Because the new flips diluted the old result. Those original 7-out-of-10 heads still happened, but they now represent a smaller fraction of the total.

Let's see this with specific numbers:

Each row adds more flips. The early "fluke" of 7 out of 10 increasingly doesn't matter. It's like adding a drop of dye to a glass of water — then pouring the glass into a bucket, and the bucket into a pool. The dye is still there, but it's imperceptible.

The key insight: Early weirdness doesn't go away — it gets overwhelmed. More data doesn't erase randomness. It drowns it out.

A Closer Look: Frequency and Proportion

Let's give names to what we've been tracking. When you flip a coin 1,000 times and get 507 heads:

• 507 is the frequency — the raw count of heads
• 0.507 is the proportion — the frequency divided by the total number of flips

In words:

$$\text{Proportion of heads} = \frac{\text{Number of heads}}{\text{Total number of flips}}$$

Frequency tells you how many times something happened. Proportion tells you how often something happened, relative to everything else. Proportion is the more revealing number, because it adjusts for scale.

Consider: "I got 507 heads" sounds like a lot. But out of 1,000 flips? That's almost exactly half. Out of 600 flips? That would be 84.5% — a wildly unfair coin. The same frequency means completely different things depending on the total.

Quick check: You roll a die 300 times and get a "6" exactly 52 times.

What is the frequency of rolling a 6? What is the proportion?
Do you think this is a fair die? Why or why not?

The frequency is 52. The proportion is $\frac{52}{300} \approx 0.173$. A fair die would give each face a proportion close to $\frac{1}{6} \approx 0.167$. So 0.173 is quite close — this looks like a fair die.

It's Not Just Coins

This pattern — individual unpredictability, collective order — isn't some quirk of coin flips. It shows up everywhere.

Try each experiment. Run them up to at least 500 trials each. Then answer:

• Do all three experiments eventually "settle down"?
• Do they settle at the same number?
• What determines where each one settles?

All three stabilize. But they stabilize at different values — 0.50, 0.167, and 0.75. The settling-down behavior is universal. Where they settle is specific to each experiment.

This is a crucial distinction: the pattern is the same (chaos → order), but the result depends on the experiment. A coin has its own number. A die has its own number. A spinner has its own number. Each random process has a kind of fingerprint — a set of proportions that it naturally settles into.

We're not ready to name that fingerprint yet. But we've just seen it. Hold onto this idea — it's the foundation for everything that follows.

What Stays the Same? What Changes?

Let's look more carefully at how different experiments compare.

What changed? What stayed the same?

Every row shows the same behavior: proportions start volatile, then stabilize. That's what stayed the same.

What changed is where they stabilize. Each experiment — and each outcome within an experiment — has its own long-run proportion.

Notice something else: the fair die gives the same proportion (0.167) for rolling a 6 and for rolling a 1. That makes sense — a fair die treats every face equally. But "rolling an even number" gives 0.50, because three out of six faces are even.

The View from Above

Let's shift perspective. Instead of watching one long sequence of flips, imagine you're watching many people each flip a coin 100 times.

Each person will get a slightly different number of heads: maybe 47, 52, 55, 48, 51, 44, 53...

Prediction: If 200 people each flip a coin 100 times and record their proportion of heads, what do you think a histogram of those 200 proportions would look like?

(a) A flat, even spread from 0 to 1
(b) All piled up at exactly 0.50
(c) Clustered around 0.50 but with some spread
(d) Two peaks, one at 0 and one at 1

The answer is (c): clustered around 0.50, with some spread. Most people land between 0.40 and 0.60. Very few land below 0.35 or above 0.65. Almost nobody gets 0.20 or 0.80.

Now try changing the number of flips per experiment:

• 10 flips each: The histogram is wide and ragged. Proportions range from 0.1 to 0.9. Lots of variability.
• 100 flips each: Narrower. Most proportions are between 0.40 and 0.60.
• 1,000 flips each: Very narrow. Nearly everyone is between 0.47 and 0.53.

What did you notice? More flips per experiment → the histogram gets tighter around 0.50. Individual results become more consistent, even though each flip is still random.

This is the same phenomenon we saw before, but from a new angle. When you increase the number of flips, the proportion doesn't just converge for one person — it converges for everyone. Randomness doesn't just settle down on average. It concentrates.

The Punchline (So Far)

Let's gather what we've discovered. No formulas, no definitions — just observations from playing with random experiments:

1. A single random outcome is genuinely unpredictable. You cannot predict the next coin flip, die roll, or spinner result.

2. The proportion of an outcome, over many trials, is surprisingly predictable. It wanders at first, then settles toward a specific number.

3. That number depends on the experiment and the outcome, not on luck or wishful thinking. A fair coin settles at 0.50 for heads. A fair die settles at 0.167 for any single face. A spinner with 3/4 blue settles at 0.75 for blue.

4. More trials → more consistency. The more times you repeat an experiment, the less the proportion wanders. The "noise" gets overwhelmed by the "signal."

5. Every random process has a pattern — a set of proportions it naturally converges to. This pattern is its fingerprint.

This is the core tension of the entire course: individual outcomes are random, but their collective behavior is structured. The rest of this course is about describing, analyzing, and using that structure.

Practice

Level 1: Concrete

Problem 1. You roll a fair six-sided die 600 times. About how many times do you expect to roll a 3?

Think about it first, then check below.

Each face of a fair die appears with a long-run proportion near $\frac{1}{6}$. Out of 600 rolls:

$$\frac{1}{6} \times 600 = 100$$

About 100 times. Not exactly 100 — that would actually be surprising — but close to it.

Problem 2. A bag contains 3 red marbles and 7 blue marbles. You draw a marble, note its color, put it back, and repeat. After 500 draws, about what proportion would you expect to be red?

The proportion of red marbles in the bag is $\frac{3}{10} = 0.30$. Over many draws, the proportion of red draws should stabilize near 0.30.

Level 2: Pattern

Problem 3. In each scenario below, predict the long-run proportion. What stays the same across them? What changes?

Work through each one before checking.

Notice: the long-run proportion always reflects the "structure" of the experiment. The coin's bias, the die's symmetry, the number of favorable outcomes relative to the total — that's what determines the long-run proportion.

Level 3: Structure

Problem 4. Your friend says: "I flipped a coin 10 times and got 8 heads. This coin must be unfair."

Do you agree? Why or why not? What if they had flipped 10,000 times and gotten 8,000 heads?

Think carefully before reading on.

With only 10 flips, getting 8 heads is unusual but not shocking — even a perfectly fair coin produces results this extreme sometimes. The proportion is 0.80, but with so few flips, proportions are noisy. We shouldn't be confident about the coin based on 10 flips.

With 10,000 flips and 8,000 heads (proportion still 0.80), the story is completely different. At that many flips, a fair coin's proportion would almost certainly be very close to 0.50. Getting 0.80 with 10,000 flips is overwhelmingly unlikely for a fair coin. Something about this coin is genuinely skewed.

The same proportion (0.80) means completely different things depending on how many trials produced it. More trials → more trustworthy proportions. This is why sample size matters — a theme that will recur throughout this course.

Level 4: Transfer

Problem 5. A hospital records whether each baby born is a boy or a girl. In a small rural clinic, about 15 babies are born each month. In a large urban hospital, about 500 babies are born each month. (Assume roughly equal chances of boy or girl.)

Which hospital is more likely to have a month where over 60% of the babies are boys?

This is the same idea as the coin flip — just in a different setting.

The small clinic. With only 15 births per month, the proportion of boys is much more volatile (just like our coin-flip proportion with few flips). Getting 60% boys out of 15 births (that's 9 or 10 boys) is well within the normal range of randomness.

The large hospital, with 500 births, would almost always see a proportion very close to 0.50. Getting over 60% boys (more than 300 out of 500) would be extraordinarily rare.

Smaller samples → more variability in proportions. This isn't just a coin-flip fact. It's a universal fact about randomness.

Debug Challenge

Problem 6. Here's a classmate's reasoning. Find the flaw:

"I flipped a coin 50 times and got 30 heads (60%). The coin is heading toward 50%, but it hasn't gotten there yet. So in my next 50 flips, I should get more tails than heads, to bring the overall proportion back to 50%. The coin 'owes' me some tails."

What's wrong with this reasoning?

This is a very common misconception. Think about it before reading on.

The flaw is the belief that the coin "remembers" its past and will "correct" for it. It won't. Each flip is completely independent — the coin has no memory. The future flips are just as likely to be 50/50 as they always were.

So how does the proportion get back to 0.50? Not by the coin compensating, but by dilution. If the next 50 flips happen to come out roughly 25 heads and 25 tails (as expected), the totals become about 55 heads out of 100 flips — proportion 0.55, closer to 0.50 than the original 0.60. No correction needed. The old imbalance just becomes a smaller fraction of a growing total.

The coin doesn't know what it's "supposed" to do. It doesn't owe you anything.

Reflection

Take a moment to think about what we've covered.

In one sentence, what is the main idea of this section?

Try to articulate it before reading on.

Here's one way to say it: Individual random outcomes are unpredictable, but the proportions of outcomes over many trials stabilize — and where they stabilize reveals the structure of the random process.

Confidence check: On a scale of 1 to 5, how confident are you that you could explain to a friend why 1,000 coin flips gives a more "reliable" proportion than 10 coin flips?

What's one thing from this section you're least sure about? Identifying the edge of your understanding is one of the most productive things you can do. If something felt fuzzy, go back and re-read that part — or experiment with the interactives again.

Creation

Design your own experiment. Think of a real-world random process — something where you can't predict individual outcomes but you could observe many of them. It could involve weather, sports, traffic, biology, games, social media — anything.

• What is one outcome you could track?
• What proportion do you think it would stabilize at, and why?
• How many observations would you need before you felt confident in that proportion?

There's no single right answer here. The goal is to start seeing the world through the lens we've been building: randomness has structure, and enough repetition reveals it.

Looking Ahead

We've discovered that random processes have a pattern — a set of proportions they converge to. But so far, we've only looked at that pattern as a single number: "heads happens about 50% of the time."

What if the random process has more than two outcomes? What if we want to see the whole pattern — not just one proportion, but all of them at once?

In Section 1.2, we'll build the tool for seeing that full picture: the histogram. It turns out that when you plot all the outcomes and their proportions, different random processes create different shapes. And those shapes are the key to everything.