Intuition The big pictureYou flip a coin n n n p p p how many successes total? That count is a Binomial random variable . The binomial distribution is just the bookkeeping of "how many ways can I get k k k
WHY it matters: any experiment that is (1) a fixed number of trials, (2) each trial success/fail, (3) constant p p p
Definition Binomial setupA random variable X X X Binomial , written X ∼ Bin ( n , p ) X \sim \text{Bin}(n,p) X ∼ Bin ( n , p )
B inary: each trial is a success/failure outcome.I ndependent: trials don't affect each other.N umber n n n fixed in advance .S ame probability p p p
Then X X X k = 0 , 1 , … , n k=0,1,\dots,n k = 0 , 1 , … , n
HOW — build one specific outcome first. Suppose the first k k k
S S ⋯ S ⏟ k F F ⋯ F ⏟ n − k \underbrace{S\,S\cdots S}_{k}\ \underbrace{F\,F\cdots F}_{n-k} k S S ⋯ S n − k F F ⋯ F Because trials are independent, multiply probabilities:
P ( this exact sequence ) = p ⋅ p ⋯ p ⏟ k ⋅ ( 1 − p ) ⋯ ( 1 − p ) ⏟ n − k = p k ( 1 − p ) n − k . P(\text{this exact sequence}) = \underbrace{p\cdot p\cdots p}_{k}\cdot\underbrace{(1-p)\cdots(1-p)}_{n-k} = p^k(1-p)^{n-k}. P ( this exact sequence ) = k p ⋅ p ⋯ p ⋅ n − k ( 1 − p ) ⋯ ( 1 − p ) = p k ( 1 − p ) n − k . Why this step? Independence lets probabilities of a joint event factor into a product.
Now count the arrangements. Any sequence with k k k same probability p k ( 1 − p ) n − k p^k(1-p)^{n-k} p k ( 1 − p ) n − k k k k n n n
( n k ) = n ! k ! ( n − k ) ! . \binom{n}{k} = \frac{n!}{k!\,(n-k)!}. ( k n ) = k ! ( n − k )! n ! . Why this step? We only care about the count k k k
Intuition WHY the mean should ben p np n p p p p p p p n n n n p np n p
HOW. Write X = X 1 + X 2 + ⋯ + X n X = X_1 + X_2 + \cdots + X_n X = X 1 + X 2 + ⋯ + X n X i = 1 X_i = 1 X i = 1 i i i 0 0 0 indicator (Bernoulli) variables.
For one indicator:
E [ X i ] = 1 ⋅ p + 0 ⋅ ( 1 − p ) = p . E[X_i] = 1\cdot p + 0\cdot(1-p) = p. E [ X i ] = 1 ⋅ p + 0 ⋅ ( 1 − p ) = p . Why this step? Expectation is ∑ ( value ) × ( prob ) \sum (\text{value})\times(\text{prob}) ∑ ( value ) × ( prob )
By linearity of expectation (works even without independence):
E [ X ] = ∑ i = 1 n E [ X i ] = n p . E[X] = \sum_{i=1}^n E[X_i] = np. E [ X ] = i = 1 ∑ n E [ X i ] = n p .
Intuition WHY independence matters hereVariance of a sum splits into individual variances only when terms are independent (covariances vanish). Trials are independent — so we get to add.
HOW. For one indicator, X i 2 = X i X_i^2 = X_i X i 2 = X i 0 2 = 0 , 1 2 = 1 0^2=0,\ 1^2=1 0 2 = 0 , 1 2 = 1 E [ X i 2 ] = p E[X_i^2]=p E [ X i 2 ] = p
Var ( X i ) = E [ X i 2 ] − ( E [ X i ] ) 2 = p − p 2 = p ( 1 − p ) . \text{Var}(X_i) = E[X_i^2] - (E[X_i])^2 = p - p^2 = p(1-p). Var ( X i ) = E [ X i 2 ] − ( E [ X i ] ) 2 = p − p 2 = p ( 1 − p ) . Why this step? Definition Var = E [ X 2 ] − ( E [ X ] ) 2 \text{Var}=E[X^2]-(E[X])^2 Var = E [ X 2 ] − ( E [ X ] ) 2 X i 2 = X i X_i^2=X_i X i 2 = X i
Independence ⇒ variances add:
Var ( X ) = ∑ i = 1 n Var ( X i ) = n p ( 1 − p ) . \text{Var}(X) = \sum_{i=1}^n \text{Var}(X_i) = np(1-p). Var ( X ) = i = 1 ∑ n Var ( X i ) = n p ( 1 − p ) .
Worked example 1 — Exactly 2 heads in 5 fair flipsn = 5 , p = 0.5 , k = 2 n=5,\ p=0.5,\ k=2 n = 5 , p = 0.5 , k = 2 P ( X = 2 ) = ( 5 2 ) ( 0.5 ) 2 ( 0.5 ) 3 = 10 ⋅ 1 32 = 10 32 = 0.3125. P(X=2)=\binom{5}{2}(0.5)^2(0.5)^3 = 10\cdot\tfrac{1}{32}=\tfrac{10}{32}=0.3125. P ( X = 2 ) = ( 2 5 ) ( 0.5 ) 2 ( 0.5 ) 3 = 10 ⋅ 32 1 = 32 10 = 0.3125. Why ( 5 2 ) = 10 \binom52=10 ( 2 5 ) = 10 Choose which 2 of 5 flips are heads. Why ( 0.5 ) 5 (0.5)^5 ( 0.5 ) 5 Each flip contributes 0.5 0.5 0.5 p = 0.5 p=0.5 p = 0.5
Worked example 2 — Defective partsA machine produces defects with p = 0.1 p=0.1 p = 0.1 n = 20 n=20 n = 20 E [ X ] = n p = 20 ( 0.1 ) = 2 defects . E[X]=np=20(0.1)=2 \text{ defects}. E [ X ] = n p = 20 ( 0.1 ) = 2 defects . Var = n p ( 1 − p ) = 20 ( 0.1 ) ( 0.9 ) = 1.8 , σ = 1.8 ≈ 1.34. \text{Var}=np(1-p)=20(0.1)(0.9)=1.8,\quad \sigma=\sqrt{1.8}\approx 1.34. Var = n p ( 1 − p ) = 20 ( 0.1 ) ( 0.9 ) = 1.8 , σ = 1.8 ≈ 1.34. Why this matters: you expect ~2 defects, typically ± 1.34 \pm1.34 ± 1.34
Worked example 3 — "At least one" (complement trick)Probability a student passes each of n = 4 n=4 n = 4 p = 0.7 p=0.7 p = 0.7 P ( at least 1 fail ) P(\text{at least 1 fail}) P ( at least 1 fail ) P ( at least 1 fail ) = 1 − P ( all pass ) P(\text{at least 1 fail}) = 1 - P(\text{all pass}) P ( at least 1 fail ) = 1 − P ( all pass ) P ( all pass ) = ( 4 4 ) ( 0.7 ) 4 ( 0.3 ) 0 = 0.7 4 = 0.2401. P(\text{all pass}) = \binom{4}{4}(0.7)^4(0.3)^0 = 0.7^4 = 0.2401. P ( all pass ) = ( 4 4 ) ( 0.7 ) 4 ( 0.3 ) 0 = 0. 7 4 = 0.2401. P ( at least 1 fail ) = 1 − 0.2401 = 0.7599. P(\text{at least 1 fail}) = 1 - 0.2401 = 0.7599. P ( at least 1 fail ) = 1 − 0.2401 = 0.7599. Why complement? "At least one" would need summing k = 0 , 1 , 2 , 3 k=0,1,2,3 k = 0 , 1 , 2 , 3
Common mistake Forgetting the( n k ) \binom{n}{k} ( k n ) Wrong idea: P ( X = k ) = p k ( 1 − p ) n − k P(X=k)=p^k(1-p)^{n-k} P ( X = k ) = p k ( 1 − p ) n − k Feels right because that's the probability of one specific sequence. Fix: many orderings give k k k ( n k ) \binom{n}{k} ( k n )
n p ( 1 − p ) np(1-p) n p ( 1 − p ) Wrong idea: mixing up mean and variance formulas. Feels right because both contain n p np n p Fix: mean = n p =np = n p ( 1 − p ) (1-p) ( 1 − p ) = n p ( 1 − p ) =np(1-p) = n p ( 1 − p ) q q q spread carries the extra factor.
Common mistake Applying binomial whenp p p Wrong idea: drawing cards without replacement and calling it binomial. Feels right because it's "success/fail." Fix: without replacement, p p p hypergeometric , not binomial. Binomial needs constant p p p
Recall Answer before revealingQ: Why does the mean have no ( 1 − p ) (1-p) ( 1 − p )
Mean comes from linearity: each trial adds p p p n p np n p p − p 2 = p ( 1 − p ) p-p^2=p(1-p) p − p 2 = p ( 1 − p ) n p ( 1 − p ) np(1-p) n p ( 1 − p ) q q q spread out outcomes are, maximal at p = 0.5 p=0.5 p = 0.5
Recall Feynman: explain to a 12-year-oldImagine tossing 10 thumbtacks. Each lands point-up with the same chance. "How many point-up?" is a binomial question. On average you get (chance × number of tacks) point-ups — that's n p np n p single tack is up doesn't change the others, so we can just add up each tack's little "wiggle" of uncertainty to get the total wiggle — that's the variance n p ( 1 − p ) np(1-p) n p ( 1 − p ) ( n k ) \binom{n}{k} ( k n )
"Mean is money you earn per try (np). Variance pays a tax of q on top (npq)."
BINS = B inary, I ndependent, N umber fixed, S ame p p p
What four conditions make a variable binomial? BINS — Binary trials, Independent, fixed Number n, Same probability p.
State the binomial PMF. P ( X = k ) = ( n k ) p k ( 1 − p ) n − k P(X=k)=\binom{n}{k}p^k(1-p)^{n-k} P ( X = k ) = ( k n ) p k ( 1 − p ) n − k for
k = 0 , … , n k=0,\dots,n k = 0 , … , n .
Where does ( n k ) \binom{n}{k} ( k n ) It counts the number of orderings of
k k k successes among
n n n trials, all equally likely.
What is E [ X ] E[X] E [ X ] Bin ( n , p ) \text{Bin}(n,p) Bin ( n , p ) n p np n p ; via
X = ∑ X i X=\sum X_i X = ∑ X i of Bernoulli indicators with
E [ X i ] = p E[X_i]=p E [ X i ] = p and linearity of expectation.
What is Var ( X ) \text{Var}(X) Var ( X ) Bin ( n , p ) \text{Bin}(n,p) Bin ( n , p ) n p ( 1 − p ) np(1-p) n p ( 1 − p ) ; each Bernoulli has variance
p ( 1 − p ) p(1-p) p ( 1 − p ) , added by independence.
Why does variance need independence but mean doesn't? Mean uses linearity (always holds); variance of a sum adds only when covariances vanish, i.e. independent trials.
For which p p p p = 0.5 p=0.5 p = 0.5 , giving
n p ⋅ 0.25 np\cdot0.25 n p ⋅ 0.25 ; zero at
p = 0 p=0 p = 0 or
1 1 1 .
Prove the PMF sums to 1. ∑ k ( n k ) p k ( 1 − p ) n − k = ( p + ( 1 − p ) ) n = 1 \sum_k\binom{n}{k}p^k(1-p)^{n-k}=(p+(1-p))^n=1 ∑ k ( k n ) p k ( 1 − p ) n − k = ( p + ( 1 − p ) ) n = 1 by the Binomial Theorem.
When is a success/fail experiment NOT binomial? When p changes or trials are dependent (e.g. sampling without replacement → hypergeometric).
Best way to compute "at least one" success? Use complement:
1 − P ( zero successes ) = 1 − ( 1 − p ) n 1-P(\text{zero successes})=1-(1-p)^n 1 − P ( zero successes ) = 1 − ( 1 − p ) n .
Bernoulli distribution — the n = 1 n=1 n = 1 Binomial Theorem — supplies ( n k ) \binom{n}{k} ( k n ) = 1 =1 = 1 Linearity of expectation — gives mean without messy sums.Variance and covariance — why independence lets variances add.Poisson distribution — limit of binomial when n → ∞ , p → 0 , n p = λ n\to\infty,\ p\to0,\ np=\lambda n → ∞ , p → 0 , n p = λ Normal approximation to binomial — for large n n n Bin ( n , p ) ≈ N ( n p , n p ( 1 − p ) ) \text{Bin}(n,p)\approx N(np,\,np(1-p)) Bin ( n , p ) ≈ N ( n p , n p ( 1 − p )) Hypergeometric distribution — the "without replacement" cousin.
One sequence prob p^k times 1-p ^n-k
Intuition Hinglish mein samjhoSocho tumhare paas ek coin hai jise tum n n n p p p Binomial random variable kehte hain. Yaad rakho BINS: B inary outcome, I ndependent trials, N umber n n n S ame probability p p p p p p
PMF ka formula ( n k ) p k ( 1 − p ) n − k \binom{n}{k}p^k(1-p)^{n-k} ( k n ) p k ( 1 − p ) n − k k k k p k ( 1 − p ) n − k p^k(1-p)^{n-k} p k ( 1 − p ) n − k ( n k ) \binom{n}{k} ( k n ) k k k
Mean n p np n p p p p n n n n p np n p linearity of expectation se aata hai, independence ki zaroorat bhi nahi. Variance n p ( 1 − p ) np(1-p) n p ( 1 − p ) p ( 1 − p ) p(1-p) p ( 1 − p ) X i 2 = X i X_i^2 = X_i X i 2 = X i independence ki wajah se sab add ho jaate hain. Dhyaan do: mean mein ( 1 − p ) (1-p) ( 1 − p ) q q q
Practical me: "at least one" type questions me hamesha complement use karo — 1 − ( 1 − p ) n 1-(1-p)^n 1 − ( 1 − p ) n
