Why Casinos Always Win: Understanding the Law of Large Numbers
Introduction: The Casino Paradox
Casinos have a problem: their customers keep winning.
No really! Walk into any casino right now and you will see slot machines paying out, blackjack players beating the dealer, someone just hit a jackpot.
Let me tell you a fun story: I went to Las Vegas long back and visited around 10 casinos in one night. My husband gave me $50 to gamble away that night because he knew those weren’t coming back! By my third casino? I was up $50! Feeling lucky, I kept going. By the end of the night, after all 10 casinos? I was roughly break-even and didn’t get any more money from my husband that night :P
But here’s the puzzle: if casinos are among the most profitable businesses in the world, and they’re continuously paying out to winners, how do they make billions?
The answer lies in the Law of Large Numbers. Once you understand it, you will see why casinos don’t need to cheat to guarantee profit. They just need to make you bet more and more (and then some more).
Law of Large Numbers: When Random Becomes Reliable
Let’s take a simple example of a coin flip. We know that both heads and tails have equal probability, which simply means that in a flip, there is equal chance of getting a head or a tail.
But what if you flip the coin 10 times? Can you be sure you’ll get exactly 5 heads and 5 tails? Probably not. Maybe you get 6 heads and 4 tails, totaling 60% heads.
You flip it 10 more times and land 6 heads and 4 tails again. Now across 20 total flips, you have 14 heads and 6tails, which is 70% heads!
You might think it’s a biased coin, so you flip it some more. After 100 flips you get 58 heads, which is 58%. Still high, but less wild than before.
After 1,000 flips you get 523 heads, which is 52.3%. After 10,000 flips you get 5,048 heads, which is 50.48%.
Notice something? As the number of trials increased, the percentage got closer and closer to 50%. Well, that’s not just luck but it’s a guarantee given by the Law of Large Numbers.
The Law of Large Numbers states that the average of the results obtained from a large number of independent random samples converges to the expected value.
But wait! What is expected value?
Let’s use the same coin flip. Say you have a bet with your friend where if the coin flip results in heads, he owes you $2, but if it results in tails, you owe him $1. You won’t make money from a single flip because that’s still random.
So instead of betting once, you decide to flip the coin 100 times.
Now let’s see how much you expect to make after 100 flips.
where:
- 0.50 is the probability of getting heads (same as getting tails)
- 100 is the number of coin flips
You get $2 richer every time you get heads but $1 poorer with tails
The result is the expected value of $50.
This means over 100 flips, you expect to win $50, which comes out to an average profit of $0.50 per flip.
Here’s the key: you will make money from this bet in the long run! Some individual flips you lose, some you win, but your average converges to that $0.50 profit per flip. The more you play, the more certain your profit becomes. That’s the Law of Large Numbers in action.
Now here’s how casinos flip this in their favor.
Casinos set up games where the expected value is negative for you and positive for them. Maybe only negative $0.05 per bet. You might win some bets, even have a lucky night. But when you and thousands of others place millions of bets, that tiny negative expected value guarantees the casino makes money.
Another application is with insurance companies, where the probability of people making a claim is much lower than people paying the premium but not claiming. So even if they have to settle claims now and then, in the long run they are sure to make money.
Central Limit Theorem: The Bell Curve Emerges from Chaos
We saw how averages converge to expected values with LLN. Now let’s see something even more interesting, which is that these averages follow a specific pattern, a bell curve.
Let me show you with dice. Each face of a die (1, 2, 3, 4, 5, 6) has equal probability of 1/6.
Take one die and roll it many times. This results in a uniform distribution, which is completely flat.
If we assign each face the value of the number on it and calculate expected value, it will be (1+2+3+4+5+6)/6 = 3.5
Now let’s do another experiment.
Roll 100 dice together and take their average. Maybe you get 4.6, or 2.3, or something around 3.5. Do this 1000 times, so each time you roll 100 dice, calculate their average, and note it down.
Now plot those 1000 averages as a histogram.
What do we see? A bell curve! There’s a peak at 3.5 and tails on both ends.
Wait, what just happened? A single die follows a uniform distribution, which is flat. But averages of 100 dice form a bell curve!
The more dice you average, the more perfect the bell curve becomes. With 10 dice averaged together, you see a hump forming. With 100 dice, you see a clear bell curve. With 1000 dice, you get a nearly perfect normal distribution.
This is the Central Limit Theorem. When you average many random things, those averages follow a bell curve, and it doesn’t matter what the original distribution was.
Started with uniform distribution like dice? You get a bell curve. Started with exponential distribution? You get a bell curve. Started with anything? You still get a bell curve.
Take averages of enough samples and a bell curve emerges. Amazing!
We see this all the time around us. If we measure the height of 1000 students in 10th grade, we will see a bell curve around the average height, with very few students being very short or very tall.
This happens because height itself is the sum of hundreds of factors like genetics, nutrition, and hormones. Each person is affected differently by these factors, but when you look at a large population, the pattern that emerges is the normal distribution. That’s CLT working at the biological level.
This means that in a nutshell, average outcomes are overwhelmingly more common than extremes, not by a little, but by an enormous factor.
LLN and CLT: Two Sides of Same Coin
You might be wondering if LLN and CLT are telling the same story. Not quite!
LLN answers: “WHERE do the averages go?” They converge to the expected value, which is the mean.
CLT answers: “HOW are the averages distributed along the way?” They follow a bell curve around that mean.
Just to recall the dice experiment above.
When a single die is rolled 1000 times, the average of all those values will be close to 3.5, which shows convergence to the expected value.
But when 100 dice are rolled together 1000 times and we plot those 1000 averages as a histogram, they form a bell curve. The curve concentrates towards the expected value and diminishes the effect of extremes.
What They Are Not!
Gambler’s Fallacy:
If you got 5 heads in a row, does LLN say tails is due? No! LLN requires that each flip is independent and does not affect upcoming flips. Each flip is still 50–50. LLN only says that over many flips, the proportion will approach 50%.
NOT Guaranteed for Small Samples:
If you flip a coin 10 times, you might get 7 heads. That’s fine. LLN and CLT need large numbers to kick in. How large? Usually 30 or more samples is the standard threshold, though more is always better.
NOT About Individual Outcomes:
LLN and CLT tell you about averages, not individual results. Individual results are random and cannot be predicted with certainty. But you can predict what happens over 10,000 flips with high confidence.
Why You Should Care
LLN and CLT are cool theorems. But why does any of this matter to you?
Because these two theorems are why statistics actually works. They are why we can make predictions about millions of people by asking just a thousand. Why factories can ensure quality without testing every product. Why casinos make billions.
Let me show you where you see these theorems in action.
Polling and Elections
We all see exit polls before election results come out. How does a poll of 1,000 people predict what 100 million voters will do? That’s LLN and CLT working together.
You sample 1,000 random voters, and their average preference (say, 52% for Candidate A) converges to the true population preference because of LLN. And because of CLT, we know that sample average follows a bell curve, which lets pollsters calculate the margin of error.
Quality Control in Manufacturing
A factory makes 10,000 bolts a day, but they cannot test every single one because it’s expensive and slow.
Instead, they sample 100 bolts, measure their lengths, and calculate the average. LLN says this average is very close to the true average of all 10,000 bolts. CLT says if the process is working correctly, measurements will form a bell curve around the target length.
If the bell curve shifts or widens, something’s wrong with the machine. They can fix it before making 10,000 bad bolts.
Medical Research
Drug trials test on maybe 1,000 people, but the drug will be used by millions. How do we know it works?
LLN says the average effect in the trial reflects the true average effect in the population. CLT says the results follow a bell curve, which lets researchers calculate confidence intervals like “we’re 95% sure the drug reduces symptoms by 20–30%.”
Without these theorems, we would need to test drugs on millions of people, which is impossible.
These theorems have many more applications in insurance, weather forecasting, A/B testing, and more.
Without these theorems, we would be guessing in the dark. With them, we turn randomness into knowledge.
Conclusion: From Uncertainty to Predictability
We started this series asking how to deal with uncertainty. Three blogs later, here’s what we’ve learned.
Blog 1 on Random Variables and Distributions: We learned to describe uncertainty. We gave it shape with bell curves, binomial, and Bernoulli distributions. We captured randomness in mathematical form.
Blog 2 on Bayes’ Theorem: We learned to update beliefs when we get evidence. We moved from skeptic to believer, rationally. We changed our minds the right way.
Blog 3 on Law of Large Numbers and Central Limit Theorem: We learned that randomness becomes predictable at scale. Individual outcomes are chaotic, but aggregate outcomes follow patterns.
See the arc? We moved from describing randomness, to reasoning about it, to predicting it.
A theme emerges here: individual events are unpredictable, but many events are remarkably predictable.
You can’t predict one coin flip, but you can predict 10,000 coin flips. You can’t predict one customer, but you can predict 100,000 customers. You can’t predict one measurement, but you can predict the average of 1,000 measurements.
This is the magic of probability and statistics. At scale, chaos becomes order. Uncertainty becomes predictability.
Why Casinos Always Win
Remember where we started? Casinos and that Las Vegas night.
Now you know the answer. They don’t win every bet. They lose lots of bets. But they have three things working in their favor:
1. Negative expected value for players, even if tiny
2. Millions of bets, which means LLN guarantees profit
3. Predictable distribution of outcomes, which means CLT lets them plan
They turned randomness into a business model. Mathematics does the work.
Our next adventure will take us to calculus, starting with derivatives. If probability taught us to handle uncertainty, calculus will teach us to handle change. Stay tuned!
