There are a surprising number of situations where probability and statistics produce results that feel deeply at odds with intuition. Many of these arise because our brains are adapted for deterministic reasoning rather than reasoning about distributions. Here are some of the most fascinating examples.
Non-transitive dice
You mentioned one of the classics.
Non-transitive dice
Imagine three or more specially numbered dice where:
- A beats B more than 50% of the time.
- B beats C more than 50% of the time.
- Yet C beats A more than 50% of the time.
It's like rock-paper-scissors, but with numbers. There is no "best" die.
Why it's strange: We instinctively expect "better than" to be transitive.
The Monty Hall problem
Monty Hall problem
Three doors:
- Pick one.
- Host opens a goat door.
- Switch?
Most people think it's now 50–50.
Actually:
Why it's strange: Opening a door isn't random—the host knows where the prize is.
Simpson's paradox
Simpson's paradox
A trend appears in multiple groups individually but reverses when the data are combined.
Example:
- Drug A works better for men.
- Drug A works better for women.
- Yet overall Drug B appears better because the groups have different sizes or difficulty levels.
This has affected real medical and admissions analyses.
The Birthday paradox
Birthday problem
Only 23 people are needed before there's over a 50% chance that two share a birthday.
At 57 people, it's over 99%.
Why it's strange: There are 365 birthdays, so people expect much larger groups.
The trick is that there are many possible pairs.
Bertrand's paradox
Bertrand paradox
What's the probability a random chord of a circle is longer than the side of the inscribed equilateral triangle?
There isn't one answer.
Depending on how "choose a random chord" is defined, you get:
Why it's strange: It shows that "random" itself requires a precise mathematical definition.
The inspection paradox
Inspection paradox
If you randomly arrive at a bus stop, you're more likely to experience a longer-than-average wait.
Why?
Long gaps occupy more time, so you're disproportionately likely to land inside one.
This also explains why:
- Friends seem to have more friends than you.
- Long songs seem to come up more often when sampling radio time.
- Hospital stays can seem longer than average when observed at random.
The friendship paradox
Friendship paradox
On average:
Your friends have more friends than you do.
This is mathematically true for most people because highly connected individuals appear in many people's friend lists.
It's been used to detect epidemics earlier by monitoring well-connected people.
The secretary problem
Secretary problem
Suppose you're interviewing 100 applicants and must hire immediately after each interview.
Optimal strategy:
- Reject the first 37%.
- Then hire the first person better than everyone you've seen.
Success probability:
- About 37%, which is actually optimal.
The mysterious number 37% comes from the constant (1/e).
Braess's paradox
Braess's paradox
Adding a new road can make everyone's commute longer.
Drivers optimize individually, producing worse global traffic.
Closing roads can sometimes improve traffic.
This has happened in real cities.
Parrondo's paradox
Parrondo's paradox
Take two games.
- Game A: lose money over time.
- Game B: lose money over time.
Alternate them in the right pattern.
Now you win.
The interaction between the games changes the overall dynamics.
The two-envelope paradox
Two-envelope paradox
You're given one of two envelopes.
One contains twice as much money as the other.
After opening yours, an expected-value argument seems to imply you should always switch.
But then you'd also always switch back.
The resolution lies in carefully modeling the prior distribution of the amounts.
The St. Petersburg paradox
St. Petersburg paradox
A coin is flipped until heads appears.
Prize:
- Heads first flip: $2
- Second flip: $4
- Third: $8
- ...
Expected value:
Infinite.
Yet almost nobody would pay anywhere near an unlimited amount to play.
This motivated ideas such as diminishing marginal utility in economics.
The gambler's fallacy
Gambler's fallacy
After ten heads in a row, many people think tails is "due."
It isn't.
For a fair coin:
- P(tails next) = 50%
- No matter what happened before.
The hot-hand effect
Hot-hand fallacy
For decades researchers believed that "being on a streak" in sports was an illusion.
More recent work suggests that, in some contexts, genuine hot-hand effects do exist, though they're often much smaller than people perceive.
It's an example of how intuition and statistical evidence can evolve with better methods.
The Sleeping Beauty problem
Sleeping Beauty problem
After a carefully designed experiment involving memory erasure, Beauty wakes up and is asked:
What probability should she assign that the coin landed heads?
Two camps argue:
Both positions are internally consistent under different interpretations of self-locating probability.
Benford's law
Benford's law
In many naturally occurring datasets:
Leading digit 1 appears about 30% of the time.
Leading digit 9 appears under 5%.
People expect digits to be equally common.
This law is used in fraud detection and forensic accounting.
Berkson's paradox
Berkson's paradox
Two unrelated traits can appear negatively correlated simply because you're only looking at a selected population.
Example:
Among students admitted to a highly selective university, exceptional math ability and exceptional writing ability might appear negatively related—not because they truly are, but because applicants with one strength can compensate for a weaker other strength during admission.
The false consensus from conditional probability
A classic example:
A disease affects 1 in 1,000 people.
A test is:
- 99% accurate for sick people.
- 99% accurate for healthy people.
You test positive.
Most people guess your chance of having the disease is around 99%.
It's actually only about 9%, because false positives vastly outnumber true positives when the disease is rare. This is an example of how base rates can dominate intuitive reasoning.
A common thread
Many of these examples fall into a few broad themes:
- Conditional information matters: Monty Hall, medical testing, Berkson's paradox.
- Sampling can be misleading: Birthday paradox, inspection paradox, friendship paradox.
- Aggregation changes conclusions: Simpson's paradox.
- Optimization can backfire: Braess's paradox, Parrondo's paradox.
- Expectation differs from typical experience: St. Petersburg paradox.
- Intuition about randomness is often unreliable: Non-transitive dice, Bertrand's paradox, gambler's fallacy.
If you're interested in even stranger territory, there are paradoxes from quantum mechanics, information theory, and decision theory—such as the Banach–Tarski paradox, Newcomb's problem, the Boy or Girl paradox, and the Blackwell–Dubins merging of opinions theorem—that push intuition even further.
