Problem 1:
What is the probability, P, a family has 2 girls, given they have 1 girl?
I first saw this problem when trying to learn about Bayes’ Theorem. “Easy!” I thought. “Given that they already have one girl… P(2 girls | 1 girl) = P(girl) = 50%.” I was shocked when I saw the answer actually is 1/3.
“Where did I go wrong?!” I spluttered. 5 minutes pass. The math was so simple, so elegant, I couldn’t possibly have made a mistake. Until it hit me.
ALREADY. “Given that they already have one girl.”
I had unwittingly made an assumption about the order of the data. I assumed they had one girl first, thus believing the only possible kid combinations were GG and GB, meaning P(2 girls | 1 girl)=50% is correct! Except, nowhere does it say the “given girl” had to be first, meaning the correct possibilities are BG, GB, GG and P(2 girls | 1 girl) = 1/3.
Lesson: Don’t implicitly assume the order of the data
Problem 2:
You are given $1 dollar and a magic coin. If you flip heads, you double the money. If it lands tails, you halve it. What is the expected amount of money after 1 flip?
This question comes from All of Statistics, a fantastic read by Dr. Wasserman.
When first approaching this problem, I figured that the probability the coin lands heads and tails is 50% each so half the time it comes up heads, the other half tails. Thus, the money would double to $2 and halve back to $1 or halve to 50 cents and then double to $1. Thus, I concluded 𝔼[money after 1 flip] = $1.
I was wrong.
The correct answer is a weighted average of the outcomes: 𝔼[d after 1 flip] = double ⋅ P(heads) + halve ⋅ P(tails) = $2 ⋅ P(heads) + $0.5 ⋅ P(tails) = $1.25
Hold up. How could the money be $1.25? The money can only double or half so it must be a multiple of $1 e.g., $0.5, $1, $2, $4. Intuitively, my logic seems sound so where did I go wrong?
It turns out, I was thinking about the “expected value” incorrectly in two ways:
The expected value actually means, “If I were to repeat this experiment infinite times, what is the average of all of those trials?” My original mistake was to keep flipping the same coin — i.e., double to $2 then halve to $1. Instead, each individual flip should be independent of the others.
Counterintuitively, the mean can be a value that is impossible for the distribution. For example, the money can never be $1.25 — it will only double to $2 or halve to $0.5, yet the “average/mean/expected value” is a weighted sum of the outcomes. Another example: in a Bernoulli distribution the result can only be 0 or 1, yet the average/expected value of the distribution is 𝔼[X] = 0.5
I still wanted to prove it to myself, so I simulated the experiment for 100,000 coin flips, getting 𝔼[X] = $1.2497 ≈ $1.25
The expected value/mean/average of an experiment means “repeat this experiment many times — what’s the average across all trials?” Also, note that each individual trial should be independent of the others.
The mean/average/expected value can be a value that’s not even in the distribution itself!
Thank you for reading!
QUESTION I: Start a game paying $1 in a pot. If Head appears the host doubles the money in the pot. If Tail appears the host takes away all the money in the pot and you need to start afresh putting $1 in the pot. What would be your strategy for the game to maximize your return?
QUESTION II: You are at zero. If head appears you go to one and if Tail appears you go to minus one. You continue your movement while tossing afresh once you complete a move. Would you ever be able to come back to zero?
REFERENCE: Two Tricky Probability Problems Blog