Skip to main content

Discrete Distributions 2

4. Poisson Distribution

4.1 Definition

A discrete random variable is said to follow the poisson distribution P(λ)\mathtt{P}(\lambda) with λR+\lambda \in\mathbb{R}_+ if:

nN,P(X=n)=λnn!eλ\forall n\in\mathbb{N},\quad \mathcal{P}(X=n)= \frac{\lambda^n}{n!}e^{-\lambda}

4.2 Significance

In probability theory and statistics, the Poisson distribution is a discrete probability distribution that expresses the probability of a given number of events occurring in a fixed interval of time or space if these events occur with a known constant mean rate and independently of the time since the last event.[1] It is named after French mathematician Siméon Denis Poisson (/ˈpwɑːsɒn/; French pronunciation: [pwasɔ̃]). The Poisson distribution can also be used for the number of events in other specified interval types such as distance, area, or volume.

For instance, a call center receives an average of 180 calls per hour, 24 hours a day. The calls are independent; receiving one does not change the probability of when the next one will arrive. The number of calls received during any minute has a Poisson probability distribution with mean 3: the most likely numbers are 2 and 3 but 1 and 4 are also likely and there is a small probability of it being as low as zero and a very small probability it could be 10.

Another example is the number of decay events that occur from a radioactive source during a defined observation period.

4.3 Moments

4.3.1 Raw Moments

nN,E[Xn]=mNmnλmm!eλ=mNmn1λm(m1)!eλ=λmN(m+1)n1λmm!eλ=λmNs=0n1(n1s)msλmm!eλ=λs=0n1(n1s)mNmsλmm!eλ=λs=0n1(n1s)E[Xs]\begin{align*} \forall n\in\mathbb{N}^*,\quad \mathbb{E}[X^n]&=\sum_{m\in\mathbb{N}}\frac{m^n\lambda^m}{m!}e^{-\lambda}\\ &=\sum_{m\in\mathbb{N}^*}\frac{m^{n-1}\lambda^m}{(m-1)!}e^{-\lambda}\\ &=\lambda\sum_{m\in\mathbb{N}}\frac{(m+1)^{n-1}\lambda^{m}}{m!}e^{-\lambda}\\ &=\lambda\sum_{m\in\mathbb{N}}\sum_{s=0}^{n-1}{n-1 \choose s}\frac{m^s\lambda^{m}}{m!}e^{-\lambda}\\ &=\lambda\sum_{s=0}^{n-1}{n-1 \choose s}\sum_{m\in\mathbb{N}}\frac{m^s\lambda^{m}}{m!}e^{-\lambda}\\ &=\lambda\sum_{s=0}^{n-1}{n-1 \choose s}\mathbb{E}[X^s] \end{align*}

In particular:

E[X]=λE[X0]=λE[1]=λ\mathbb{E}[X]=\lambda\mathbb{E}[X^0]=\lambda\mathbb{E}[1]=\lambda

4.3.2 Central Moments

We will start by the variance:

E[X2]=λs=01(1s)E[Xs]=λE[X0]+λE[X1]=λ+λ2    V[X]=E[X2]E[X]2=λ\begin{align*} \mathbb{E}[X^2]&=\lambda \sum_{s=0}^1{1 \choose s}\mathbb{E}[X^s]\\ &=\lambda\mathbb{E}[X^0]+\lambda \mathbb{E}[X^1]\\ &=\lambda +\lambda^2\\ \implies \mathbb{V}[X]&=\mathbb{E}[X^2]-\mathbb{E}[X]^2\\ &=\lambda \end{align*}

5. Hyper-geometric Distribution

5.1 Prelude

For a random variable XX and event EE.

We will denote by X[E]X[E] the random variable XX knowing that EE occured

5.2 Definition

Let N,K,nN/KN and nNN,K,n\in\mathbb{N}/ K\le N \space\text{and}\space n\le N

We say that a random variable XX follows the hyper-geometric distribution H(n,N,K)\mathcal{H}(n,N,K) with parameters N,K,nN,K,n if:

X1,,XNB(p) i.i.d /X=Sn[SN=K]with Sk=i=1kXik{1,,N}\exists X_1,\dots,X_{N} \sim \mathcal{B}(p) \space \text{i.i.d}\space / \quad X=S_{n}\left[S_N=K\right] \quad \text{with}\space S_k=\sum_{i=1}^{k}X_i \quad \forall k\in\{1,\dots,N\}

5.2 Significance

In probability theory and statistics, the hyper-geometric distribution is a discrete probability distribution that describes the probability of kk successes (random draws for which the object drawn has a specified feature) in nn draws, without replacement, from a finite population of size NN that contains exactly KK objects with that feature, wherein each draw is either a success or a failure.

In contrast, the binomial distribution describes the probability of successes in nn draws with replacement.

5.3 Probability Mass function

We have Sn=i=1nXiB(n,p)S_n=\sum_{i=1}^n X_i\sim \mathcal{B}(n,p) and SNSn=i=n+1NXiB(Nn,p)S_N-S_n=\sum_{i=n+1}^N X_i \sim \mathcal{B}(N-n,p).

In fact, SnS_n and SNSnS_{N}-S_n are independent.

With that we have:

k{0,,K},P(X=k)=P(Sn=kSN=K)=P(Sn=k  SN=K)P(SN=K)=P(Sn=k  SNSn=Kk)P(SN=K)=P(Sn=k)P(SNSn=Kk)P(SN=K)=(nk)pk(1p)nk×(NnKk)pKk(1p)NnK+k(NK)pK(1p)NK=(nk)×(NnKk)(NK)=n!(Nn)!K!(NK)!k!(nk)!(Kk)!(NnK+k)!N!=K!k!(Kk)!(NK)!(nk)!(Nnn+k)!n!(Nn)!N!=(Kk)(NKnk)(Nn)\begin{align*} \forall k\in\{0,\dots,K\},\quad \mathcal{P}(X=k)&=\mathcal{P}(S_n=k \mid S_N=K) \\ &= \frac{\mathcal{P}(S_n=k\space \wedge \space S_N=K)}{\mathcal{P}(S_N=K)}\\ &= \frac{\mathcal{P}(S_n=k\space \wedge \space S_N-S_n=K-k)}{\mathcal{P}(S_N=K)} \\ &=\frac{\mathcal{P}(S_n=k)\cdot \mathcal{P}(S_N-S_n=K-k)}{\mathcal{P}(S_N=K)} \\ &= \frac{{n \choose k}p^k(1-p)^{n-k}\times {N-n \choose K-k}p^{K-k}(1-p)^{N-n-K+k}}{{N \choose K}p^K(1-p)^{N-K}} \\ &= \frac{{n \choose k} \times {N-n \choose K-k} }{{N \choose K}} \\ &= \frac{n!(N-n)!K!(N-K)!}{k!(n-k)!(K-k)!(N-n-K+k)!N!}\\ &= \frac{K!}{k!(K-k)!}\cdot \frac{(N-K)!}{(n-k)!(N-n-n+k)!}\cdot \frac{n!(N-n)!}{N!}\\ &= \frac{{K \choose k}\cdot {N-K \choose n-k}}{{N \choose n}} \end{align*}

5.4 Moments

5.4.1 Raw Moments

We will start by the expected value E[X]:\mathbb{E}[X]:

E[X]=E[SnSN=K]=k=1nE[XkSN=K]=k=1n P(Xk=1SN=K)=k=1nk=1n P(Xk=1SN=K)P(SN=K)=k=1nk=1n P(Xk=1SNXk=K1)P(SN=K)=k=1nP(Xk=1)P(SNXk=K1)P(SN=K)as SNXk and Xk are independent=k=1np(N1K1)pK1(1p)NK(NK)pK(1P)NK=k=1n(N1K1)(NK)=n(N1)!K!(NK)!K!(NK)!N!=nNK\begin{align*} \mathbb{E}[X]&=\mathbb{E}[S_n \mid S_N= K] \\ &= \sum_{k=1}^n\mathbb{E}[X_k \mid S_N=K] \\ &= \sum_{k=1}^n\ \mathcal{P}(X_k = 1 \mid S_N = K) \\ &= \sum_{k=1}^n \frac{\sum_{k=1}^n\ \mathcal{P}(X_k = 1 \wedge S_N = K)}{\mathcal{P}( S_N = K)} \\ &= \sum_{k=1}^n \frac{\sum_{k=1}^n\ \mathcal{P}(X_k = 1 \wedge S_N-X_k = K-1)}{\mathcal{P}( S_N = K)} \\ &=\sum_{k=1}^n \frac{\mathcal{P}(X_k = 1) \cdot \mathcal{P}(S_N-X_k = K-1)}{\mathcal{P}( S_N = K)} \quad \text{as}\space S_N-X_k\space \text{and} \space X_k \space \text{are independent} \\ &= \sum_{k=1}^n \frac{p \cdot {N-1 \choose K-1}p^{K-1} (1-p)^{N-K}}{{N \choose K} p^K (1-P)^{N-K}} \\ &= \sum_{k=1}^n \frac{{N-1 \choose K-1}}{{N \choose K}} \\ &= n\cdot \frac{(N-1)!K!(N-K)!}{K!(N-K)!N!}\\ &= n\frac{N}{K} \end{align*}

5.4.2 Central Moments

For convenience, we will denote Yi=Xi[Sn=K]Y_i=X_i[S_n=K]

We will start by the variance V[X]:\mathbb{V}[X]:

V[X]=Cov(Sn[SN=K],Sn[SN=K])=Cov((i=1nXi)[SN=K],(j=1nXj)[SN=K])=Cov(i=1nXi[SN=K],j=1nXj[SN=K])=i=1nV[Yi]+21i<jnCov(Yi,Yj)ij,Cov(Yi,Yj)=E[Yi×Yj]E[Yi]×E[Yj]=P[(Yi×Yj)=1]K2N2=P((Yi×Yj)=1Yj=1)×P(Yj=1)K2N2=P(Yi=1Yj=1)×P(Yj=1)K2N2=P(Yi=1Yj=1SN=K)×KNK2N2=P(Xi=1Xj=1SN=K)×KNK2N2=P(Xi=1Xj=1SN=K)P(Xj=1Sn=K)×KNK2N2=P(Xi=1Xj=1SNXiXj=K2)P(Xj=1SNXj=K1)×KNK2N2=P(Xi=1)P(Xj=1)P(SNXiXj=K2)P(Xj=1)P(SNXj=K1)×KNK2N2=pp(N2K2)pK2(1p)NKp(N1K1)pK1(1p)NK×KNK2N2=K1N1×KNK2N2=KN(K1N1KN)    V[X]=i=1nV[Yi]+21i<jnCov(Yi,Yj)=nKNNKN+2×n(n1)2KN(K1N1KN)=KN(nNKN+n(n1)(K1N1KN))=nKN(NKN+(n1)(K1N1KN))=nKN((NK)(N1)+N(n1)(K1)K(n1)(N1)N(N1))=nKN(N2KNN+K+nNKNKnN+NnNK+KN+nKKN(N1))=nKN(N2NKnN+nKN(N1))=nKN(N(NK)n(NK)N(N1))=nKNNKNNnN1\begin{align*} \mathbb{V}[X]&= \text{Cov}\left(S_n[S_N=K],S_n[S_N=K]\right) \\ &= \text{Cov}\left(\left(\sum_{i=1}^n X_i\right)\left[S_N=K\right],\left(\sum_{j=1}^nX_j\right)\left[S_N=K\right]\right) \\ &= \text{Cov}\left(\sum_{i=1}^n X_i [S_N=K],\sum_{j=1}^nX_j [S_N=K]\right) \\ &= \sum_{i=1}^n \mathbb{V}[Y_i]+2\sum_{1\le i<j\le n }\text{Cov} \left(Y_i,Y_j\right)\\ \forall i\neq j,\quad\text{Cov}(Y_i,Y_j)&=\mathbb{E}[Y_i \times Y_j]-\mathbb{E}[Y_i]\times \mathbb{E}[Y_j] \\ &= \mathcal{P}\left[\left(Y_i\times Y_j\right) =1\right] -\frac{K^2}{N^2}\\ &= \mathcal{P}\left(\left(Y_i\times Y_j\right) = 1 \mid Y_j=1\right)\times \mathcal{P}(Y_j=1)-\frac{K^2}{N^2} \\ &= \mathcal{P}\left(Y_i = 1 \mid Y_j=1\right)\times \mathcal{P}(Y_j=1)-\frac{K^2}{N^2} \\ &= \mathcal{P}\left(Y_i=1 \mid Y_j=1\wedge S_N=K\right)\times \frac{K}{N}-\frac{K^2}{N^2} \\ &= \mathcal{P}\left(X_i=1 \mid X_j=1\wedge S_N=K\right)\times \frac{K}{N}-\frac{K^2}{N^2} \\ &= \frac{\mathcal{P}\left(X_i=1 \wedge X_j=1\wedge S_N=K\right)}{\mathcal{P}(X_j=1\wedge S_n=K)}\times \frac{K}{N}-\frac{K^2}{N^2} \\ &=\frac{\mathcal{P}\left(X_i=1 \wedge X_j=1\wedge S_N-X_i-X_j=K-2\right)}{\mathcal{P}(X_j=1\wedge S_N-X_j=K-1)}\times \frac{K}{N}-\frac{K^2}{N^2} \\ &=\frac{\mathcal{P}\left(X_i=1) \cdot \mathcal{P}(X_j=1) \cdot \mathcal{P}( S_N-X_i-X_j=K-2\right)}{\mathcal{P}(X_j=1)\cdot \mathcal{P}( S_N-X_j=K-1)}\times \frac{K}{N}-\frac{K^2}{N^2} \\ &=\frac{p\cdot p \cdot {N-2 \choose K-2}p^{K-2}(1-p)^{N-K}}{p\cdot {N-1 \choose K-1}p^{K-1}(1-p)^{N-K}}\times \frac{K}{N}-\frac{K^2}{N^2} \\ &= \frac{K-1}{N-1}\times \frac{K}{N}-\frac{K^2}{N^2}\\ &= \frac{K}{N}\left(\frac{K-1}{N-1}-\frac{K}{N}\right)\\ \implies \mathbb{V}[X]&= \sum_{i=1}^n \mathbb{V}[Y_i]+2\sum_{1\le i<j\le n }\text{Cov} \left(Y_i,Y_j\right)\\ &= n\frac{K}{N}\frac{N-K}{N}+2\times \frac{n(n-1)}{2}\cdot \frac{K}{N}\left(\frac{K-1}{N-1}-\frac{K}{N}\right) \\ &=\frac{K}{N}\left(n\frac{N-K}{N}+n(n-1)\cdot \left( \frac{K-1}{N-1}-\frac{K}{N}\right)\right) \\ &= n\frac{K}{N}\left(\frac{N-K}{N}+(n-1)\cdot \left( \frac{K-1}{N-1}-\frac{K}{N}\right)\right) \\ &= n\frac{K}{N}\left(\frac{(N-K)(N-1)+N(n-1)(K-1)-K(n-1)(N-1)}{N(N-1)}\right) \\ &= n\frac{K}{N}\left(\frac{N^2-KN-N+K+nNK-NK-nN+N-nNK+KN+nK-K}{N(N-1)}\right) \\ &=n\frac{K}{N}\left(\frac{N^2-NK-nN+nK}{N(N-1)}\right)\\ &=n\frac{K}{N}\left(\frac{N(N-K)-n(N-K)}{N(N-1)}\right)\\ &=n\frac{K}{N}\cdot \frac{N-K}{N}\cdot \frac{N-n}{N-1} \end{align*}