Special Discrete Random Variables

Discrete Uniform Random Variable

  • Parameters and Support:

    • Defined by two integers aa and bb satisfying a≤ba \le b.
    • The discrete uniform random variable X∼DUnif(a,b)X \sim \text{DUnif}(a, b) models an experiment where outcomes take on integer values between aa and bb inclusive.
    • The support of XX consists of b−a+1b - a + 1 consecutive integer values: {a,a+1,a+2,…,b−1,b}\{a, a + 1, a + 2, \dots, b-1, b\}.
    • Every possible outcome within this range is equally likely.
  • Probability Mass Function (PMF):

    • The PMF assigns equal probability to every integer in the support:     pX(x)=1b−a+1for x∈{a,a+1,…,b}p_X(x) = \frac{1}{b - a + 1} \quad \text{for } x \in \{a, a + 1, \dots, b\}
    • For any value outside this set, pX(x)=0p_X(x) = 0.

Probability mass function of the discrete uniform distribution

  • Expected Value and Variance:
    • The expectation represents the arithmetic midpoint of the support interval:     E(X)=a+b2E(X) = \frac{a + b}{2}
    • The variance measures the dispersion around the mean:     Var(X)=(b−a+1)2−112\text{Var}(X) = \frac{(b - a + 1)^2 - 1}{12}

Bernoulli and Binomial Random Variables

  • The Bernoulli Random Variable:

    • Describes a single trial with exactly two possible outcomes: Success (SS) and Failure (FF).
    • The parameter is p∈[0,1]p \in [0, 1], denoting the probability of success: P(S)=pP(S) = p, and P(F)=1−pP(F) = 1 - p.
    • Often utilized as an indicator random variable, mapped numerically as:     x=1if Sx = 1 \quad \text{if } Sx=0if Fx = 0 \quad \text{if } F
    • The tabular probability mass function is:
    • pX(0)=1−pp_X(0) = 1 - p
    • pX(1)=pp_X(1) = p
    • The compact algebraic definition of the PMF for X∼Ber(p)X \sim \text{Ber}(p) is:     pX(x)=px(1−p)1−xwhere x∈{0,1}p_X(x) = p^x (1 - p)^{1 - x} \quad \text{where } x \in \{0, 1\}
    • Expectation and variance for a Bernoulli random variable:     E(X)=0⋅(1−p)+1⋅p=pE(X) = 0 \cdot (1 - p) + 1 \cdot p = pE(X2)=02⋅(1−p)+12⋅p=pE(X^2) = 0^2 \cdot (1 - p) + 1^2 \cdot p = pVar(X)=E(X2)−[E(X)]2=p−p2=p(1−p)\text{Var}(X) = E(X^2) - [E(X)]^2 = p - p^2 = p(1 - p)
  • The Binomial Experiment Framework:

    • Consists of nn identical, independent trials.
    • Each individual trial results in either Success (SS) with probability pp, or Failure (FF) with probability 1−p1 - p.
    • The sample space Ω\Omega consists of all possible sequences of SS and FF of length nn, containing 2n2^n outcomes.
    • The Binomial random variable XX counts the total number of successes in nn trials: X=number of successesX = \text{number of successes}.
    • Support of XX: x∈{0,1,2,…,n}x \in \{0, 1, 2, \dots, n\}.
  • Derivation of the Binomial PMF (n=3n = 3 Case Study):

    • For n=3n = 3, the sample space Ω\Omega partitions according to the number of successes XX:
    • X=0X = 0: Outcome {FFF}\{FFF\}. Probability is P(FFF)=(1−p)3=(30)p0(1−p)3−0P(FFF) = (1-p)^3 = \binom{3}{0} p^0 (1-p)^{3-0}.
    • X=1X = 1: Outcomes {SFF,FSF,FFS}\{SFF, FSF, FFS\}. Each sequence has probability p(1−p)2p(1-p)^2. Total probability is PX(1)=3p(1−p)2=(31)p1(1−p)3−1P_X(1) = 3 p (1-p)^2 = \binom{3}{1} p^1 (1-p)^{3-1}.
    • X=2X = 2: Outcomes {SSF,SFS,FSS}\{SSF, SFS, FSS\}. Each sequence has probability p2(1−p)p^2 (1-p). Total probability is PX(2)=P(SSF)+P(SFS)+P(FSS)=3p2(1−p)=(32)p2(1−p)3−2P_X(2) = P(SSF) + P(SFS) + P(FSS) = 3 p^2 (1-p) = \binom{3}{2} p^2 (1-p)^{3-2}.
    • X=3X = 3: Outcome {SSS}\{SSS\}. Probability is P(SSS)=p3=(33)p3(1−p)3−3P(SSS) = p^3 = \binom{3}{3} p^3 (1-p)^{3-3}.
    • Extending to general nn and xx successes, there are (nx)\binom{n}{x} mutually exclusive sequences, each having probability px(1−p)n−xp^x (1 - p)^{n - x}.
    • General Definition of the PMF for X∼Bin(n,p)X \sim \text{Bin}(n, p):     pX(x)=(nx)px(1−p)n−xwhere x=0,1,…,np_X(x) = \binom{n}{x} p^x (1 - p)^{n - x} \quad \text{where } x = 0, 1, \dots, n
  • Representation as a Sum of Independent Bernoulli Trials:

    • Let X1,X2,…,XnX_1, X_2, \dots, X_n be independent and identically distributed (i.i.d.) Bernoulli random variables such that Xi∼Ber(p)X_i \sim \text{Ber}(p).
    • Each XiX_i indicates the outcome of trial ii (Xi=1X_i = 1 for success, Xi=0X_i = 0 for failure).
    • The Binomial random variable can be written directly as their sum:     X=X1+X2+⋯+Xn=∑i=1nXiX = X_1 + X_2 + \dots + X_n = \sum_{i=1}^n X_i
    • Tabular correspondence between Ω\Omega outcomes and Bernoulli components (n=3n = 3):
    • Outcome FFF  ⟹  X1=0,X2=0,X3=0  ⟹  X=0FFF \implies X_1 = 0, X_2 = 0, X_3 = 0 \implies X = 0
    • Outcomes SFF,FSF,FFS  ⟹  SFF, FSF, FFS \implies each has one 11 and two 00s   ⟹  X=1\implies X = 1
    • Outcomes SSF,SFS,FSS  ⟹  SSF, SFS, FSS \implies each has two 11s and one 00   ⟹  X=2\implies X = 2
    • Outcome SSS  ⟹  X1=1,X2=1,X3=1  ⟹  X=3SSS \implies X_1 = 1, X_2 = 1, X_3 = 1 \implies X = 3
  • Moment Generating Function (MGF) of the Binomial Distribution:

    • The Binomial Theorem states:     ∑x=0n(nx)axbn−x=(a+b)n\sum_{x=0}^n \binom{n}{x} a^x b^{n-x} = (a + b)^n
    • Derivation of the MGF ϕX(t)=E[etX]\phi_X(t) = E[e^{tX}]:     ϕX(t)=∑x=0netxpX(x)\phi_X(t) = \sum_{x=0}^n e^{tx} p_X(x)ϕX(t)=∑x=0netx(nx)px(1−p)n−x\phi_X(t) = \sum_{x=0}^n e^{tx} \binom{n}{x} p^x (1 - p)^{n - x}ϕX(t)=∑x=0n(nx)(pet)x(1−p)n−x\phi_X(t) = \sum_{x=0}^n \binom{n}{x} (p e^t)^x (1 - p)^{n - x}
    • Setting a=peta = p e^t and b=1−pb = 1 - p in the Binomial Theorem gives:     ϕX(t)=(pet+1−p)n\phi_X(t) = (p e^t + 1 - p)^n
    • Evaluating at t=0t = 0 verifies standard MGF normalization:     ϕX(0)=(pe0+1−p)n=(p+1−p)n=1n=1\phi_X(0) = (p e^0 + 1 - p)^n = (p + 1 - p)^n = 1^n = 1
  • Derivation of Mean and Variance:

    • Using the MGF first derivative:     ϕX′(t)=n(pet+1−p)n−1⋅pet\phi_X'(t) = n (p e^t + 1 - p)^{n-1} \cdot p e^tE(X)=ϕX′(0)=n(p+1−p)n−1p=npE(X) = \phi_X'(0) = n (p + 1 - p)^{n-1} p = n p
    • Using the MGF second derivative:     ϕX′′(t)=n(n−1)(pet+1−p)n−2(pet)2+n(pet+1−p)n−1pet\phi_X''(t) = n(n - 1)(p e^t + 1 - p)^{n-2} (p e^t)^2 + n(p e^t + 1 - p)^{n-1} p e^tE(X2)=ϕX′′(0)=n(n−1)p2+np=n2p2−np2+npE(X^2) = \phi_X''(0) = n(n - 1)p^2 + np = n^2 p^2 - n p^2 + n p
    • Calculating the variance:     Var(X)=E(X2)−[E(X)]2=(n2p2−np2+np)−(np)2=np(1−p)\text{Var}(X) = E(X^2) - [E(X)]^2 = (n^2 p^2 - n p^2 + n p) - (n p)^2 = n p (1 - p)
    • Alternatively, by the linear combination of i.i.d. Bernoulli variables:     E(X)=E(∑i=1nXi)=∑i=1nE(Xi)=∑i=1np=npE(X) = E\left(\sum_{i=1}^n X_i\right) = \sum_{i=1}^n E(X_i) = \sum_{i=1}^n p = n pVar(X)=Var(∑i=1nXi)=∑i=1nVar(Xi)=∑i=1np(1−p)=np(1−p)\text{Var}(X) = \text{Var}\left(\sum_{i=1}^n X_i\right) = \sum_{i=1}^n \text{Var}(X_i) = \sum_{i=1}^n p(1 - p) = n p(1 - p)
  • Application Example: Multiple-Choice Exam Guessing:

    • An exam has n=10n = 10 questions, each with 44 alternatives (one correct answer per question).
    • An unprepared student guesses all answers completely at random, so p=14=0.25p = \frac{1}{4} = 0.25.
    • Model: X∼Bin(10,0.25)X \sim \text{Bin}(10, 0.25), where XX is the number of correct answers.
    • Probability of getting exactly 88 correct answers:     P(X=8)=pX(8)=(108)(14)8(1−14)10−8=(108)(14)8(34)2P(X = 8) = p_X(8) = \binom{10}{8} \left(\frac{1}{4}\right)^8 \left(1 - \frac{1}{4}\right)^{10 - 8} = \binom{10}{8} \left(\frac{1}{4}\right)^8 \left(\frac{3}{4}\right)^2
    • Expected number of correct answers:     E(X)=np=10×14=2.5E(X) = n p = 10 \times \frac{1}{4} = 2.5
    • Variance of correct answers:     Var(X)=np(1−p)=10×14×34=3016=1.875≈1.9\text{Var}(X) = n p (1 - p) = 10 \times \frac{1}{4} \times \frac{3}{4} = \frac{30}{16} = 1.875 \approx 1.9
  • Distribution Shape and Symmetry:

    • Skewness is governed by the success probability pp:
    • p<0.5p < 0.5: The distribution is right-skewed (positively skewed).
    • p=0.5p = 0.5: The distribution is perfectly symmetric around the mean E(X)=n2E(X) = \frac{n}{2}.
    • p>0.5p > 0.5: The distribution is left-skewed (negatively skewed).
    • Shape comparisons for n=10n = 10:
    • p=0.25  ⟹  E(X)=2.5p = 0.25 \implies E(X) = 2.5, Var(X)=1.875≈1.9\text{Var}(X) = 1.875 \approx 1.9 (right-skewed).
    • p=0.50  ⟹  E(X)=5.0p = 0.50 \implies E(X) = 5.0, Var(X)=2.5\text{Var}(X) = 2.5 (symmetric).
    • p=0.75  ⟹  E(X)=7.5p = 0.75 \implies E(X) = 7.5, Var(X)=1.875≈1.9\text{Var}(X) = 1.875 \approx 1.9 (left-skewed).

Binomial distribution PMF plots for n = 10 across p = 0.25, 0.5, and 0.75

  • Shape comparisons for n=50n = 50:
    • p=0.25  ⟹  E(X)=12.5p = 0.25 \implies E(X) = 12.5, Var(X)=9.375≈9.4\text{Var}(X) = 9.375 \approx 9.4.
    • p=0.50  ⟹  E(X)=25.0p = 0.50 \implies E(X) = 25.0, Var(X)=12.5\text{Var}(X) = 12.5 (symmetric, bell-shaped).
    • p=0.75  ⟹  E(X)=37.5p = 0.75 \implies E(X) = 37.5, Var(X)=9.375≈9.4\text{Var}(X) = 9.375 \approx 9.4.

Binomial distribution PMF plots for n = 50 across p = 0.25, 0.5, and 0.75

Poisson Random Variable

  • Definition and Context:

    • Models the number of occurrences of an event over a specified period of time or region of space.
    • Governed by rate parameter λ>0\lambda > 0, which denotes the average number of occurrences per measurement interval: λ=np\lambda = n p.
    • Canonical real-world processes:
    • Number of phone calls received by a server per hour.
    • Number of laser photons hitting a detector per second.
    • Number of spam emails arriving in an inbox per day.
  • Mathematical Derivation from the Binomial PMF (Poisson Limit):

    • Arises as the limiting form of the Binomial distribution under the following conditions:
    • The number of trials grows infinitely large: n→∞n \rightarrow \infty
    • The success probability per trial becomes infinitesimally small: p→0p \rightarrow 0
    • The product remains constant: λ=np\lambda = n p
    • Derivation of the limit:     pX(x)=lim⁡n→∞(nx)px(1−p)n−x=lim⁡n→∞n(n−1)⋯(n−x+1)x!(λn)x(1−λn)n−xp_X(x) = \lim_{n \to \infty} \binom{n}{x} p^x (1 - p)^{n - x} = \lim_{n \to \infty} \frac{n(n-1)\cdots(n-x+1)}{x!} \left(\frac{\lambda}{n}\right)^x \left(1 - \frac{\lambda}{n}\right)^{n - x}pX(x)=λxx!lim⁡n→∞[n(n−1)⋯(n−x+1)nx(1−λn)n(1−λn)−x]p_X(x) = \frac{\lambda^x}{x!} \lim_{n \to \infty} \left[ \frac{n(n-1)\cdots(n-x+1)}{n^x} \left(1 - \frac{\lambda}{n}\right)^n \left(1 - \frac{\lambda}{n}\right)^{-x} \right]
    • Since lim⁡n→∞n(n−1)⋯(n−x+1)nx=1\lim_{n \to \infty} \frac{n(n-1)\cdots(n-x+1)}{n^x} = 1, lim⁡n→∞(1−λn)n=e−λ\lim_{n \to \infty} \left(1 - \frac{\lambda}{n}\right)^n = e^{-\lambda}, and lim⁡n→∞(1−λn)−x=1\lim_{n \to \infty} \left(1 - \frac{\lambda}{n}\right)^{-x} = 1:     pX(x)=e−λλxx!where x=0,1,2,…p_X(x) = \frac{e^{-\lambda} \lambda^x}{x!} \quad \text{where } x = 0, 1, 2, \dots
  • Poisson Approximation to the Binomial:

    • When nn is large and pp is small, the Binomial PMF is approximated as:     (nx)px(1−p)n−x≈e−np(np)xx!\binom{n}{x} p^x (1 - p)^{n - x} \approx \frac{e^{-n p} (n p)^x}{x!}
    • Example comparison for X∼Bin(20,0.02)X \sim \text{Bin}(20, 0.02), computing P(X=2)P(X = 2), where n=20n = 20 and p=0.02p = 0.02:
    • a) Exact Binomial evaluation:       P(X=2)=(202)(0.02)2(1−0.02)20−2=190×(0.0004)×(0.98)18≈0.0528P(X = 2) = \binom{20}{2} (0.02)^2 (1 - 0.02)^{20 - 2} = 190 \times (0.0004) \times (0.98)^{18} \approx 0.0528
    • b) Poisson approximation evaluation (λ=np=20×0.02=0.4\lambda = n p = 20 \times 0.02 = 0.4):       P(X=2)≈e−0.4(0.4)22!=e−0.4×0.162≈0.0536P(X = 2) \approx \frac{e^{-0.4} (0.4)^2}{2!} = \frac{e^{-0.4} \times 0.16}{2} \approx 0.0536
  • Moment Generating Function (MGF) of the Poisson Distribution:

    • Maclaurin series expansion of the exponential function:     ∑x=0∞bxx!=eb\sum_{x=0}^\infty \frac{b^x}{x!} = e^b
    • Derivation of ϕX(t)=E[etX]\phi_X(t) = E[e^{tX}]:     ϕX(t)=∑x=0∞etxpX(x)=∑x=0∞etxe−λλxx!=e−λ∑x=0∞(λet)xx!\phi_X(t) = \sum_{x=0}^\infty e^{tx} p_X(x) = \sum_{x=0}^\infty e^{tx} \frac{e^{-\lambda} \lambda^x}{x!} = e^{-\lambda} \sum_{x=0}^\infty \frac{(\lambda e^t)^x}{x!}
    • Substituting b=λetb = \lambda e^t into the Maclaurin expansion yields:     ϕX(t)=e−λeλet=exp⁡(λ(et−1))\phi_X(t) = e^{-\lambda} e^{\lambda e^t} = \exp(\lambda(e^t - 1))
  • Expectation and Variance:

    • From the MGF:     ϕX′(t)=λetexp⁡(λ(et−1))  ⟹  E(X)=ϕX′(0)=λ\phi_X'(t) = \lambda e^t \exp(\lambda(e^t - 1)) \implies E(X) = \phi_X'(0) = \lambdaϕX′′(t)=λetexp⁡(λ(et−1))+(λet)2exp⁡(λ(et−1))  ⟹  E(X2)=ϕX′′(0)=λ+λ2\phi_X''(t) = \lambda e^t \exp(\lambda(e^t - 1)) + (\lambda e^t)^2 \exp(\lambda(e^t - 1)) \implies E(X^2) = \phi_X''(0) = \lambda + \lambda^2Var(X)=E(X2)−[E(X)]2=(λ+λ2)−λ2=λ\text{Var}(X) = E(X^2) - [E(X)]^2 = (\lambda + \lambda^2) - \lambda^2 = \lambda
    • A distinctive property of the Poisson distribution is that its mean and variance are identical: E(X)=Var(X)=λE(X) = \text{Var}(X) = \lambda.
  • Application Example: Traffic Accident Rate:

    • Suppose accidents occur at an average rate of 1414 accidents per week.
    • Probability of at least one accident in a week (X∼Poisson(14)X \sim \text{Poisson}(14)):     P(X≥1)=1−P(X=0)=1−e−141400!=1−e−14≈1−8.315×10−7≈0.999999P(X \ge 1) = 1 - P(X = 0) = 1 - \frac{e^{-14} 14^0}{0!} = 1 - e^{-14} \approx 1 - 8.315 \times 10^{-7} \approx 0.999999
    • Probability of at least one accident today:
    • The daily rate is scaled linearly: λday=147=2\lambda_{\text{day}} = \frac{14}{7} = 2 accidents per day.
    • Let Y∼Poisson(2)Y \sim \text{Poisson}(2) represent daily accidents:       P(Y≥1)=1−P(Y=0)=1−e−2200!=1−e−2≈1−0.1353=0.8647P(Y \ge 1) = 1 - P(Y = 0) = 1 - \frac{e^{-2} 2^0}{0!} = 1 - e^{-2} \approx 1 - 0.1353 = 0.8647
  • Visual Behavior across Parameter Values:

    • For small rates (λ=2\lambda = 2): The distribution is strongly right-skewed with a peak at small integer counts (x=1,2x = 1, 2), with E(Y)=2E(Y) = 2 and Var(Y)=2\text{Var}(Y) = 2.
    • For larger rates (λ=14\lambda = 14): The distribution shifts rightward, becomes substantially more symmetric, and approximates a continuous Gaussian bell curve centered at E(X)=14E(X) = 14 with Var(X)=14\text{Var}(X) = 14.

Poisson distribution PMF comparison for lambda = 14 and lambda = 2

Geometric Random Variable

  • Definition and Experiment Framework:

    • Formulated upon an infinite sequence of independent, identical Bernoulli trials.
    • Each trial has two possible outcomes, Success (SS) and Failure (FF), with constant success probability P(S)=pP(S) = p.
    • The random variable XX counts the total number of trials required up to and including the first occurrence of success.
    • Support of XX: x∈{1,2,3,… }x \in \{1, 2, 3, \dots\}.
  • Derivation of the Geometric PMF:

    • The event X=xX = x requires that the first x−1x - 1 trials end in failure, and trial xx ends in success (FF…F⏟x−1 timesS\underbrace{F F \dots F}_{x-1 \text{ times}} S).
    • Because trials are independent, probabilities multiply directly:     pX(x)=(1−p)x−1pwhere x=1,2,3,…p_X(x) = (1 - p)^{x - 1} p \quad \text{where } x = 1, 2, 3, \dots
  • Moment Generating Function (MGF) of the Geometric Distribution:

    • Geometric series summation formula:     ∑x=1∞bx=b1−bfor ∣b∣<1\sum_{x=1}^\infty b^x = \frac{b}{1 - b} \quad \text{for } |b| < 1
    • Derivation of ϕX(t)=E[etX]\phi_X(t) = E[e^{tX}]:     ϕX(t)=∑x=1∞etxpX(x)=∑x=1∞etx(1−p)x−1p=p1−p∑x=1∞((1−p)et)x\phi_X(t) = \sum_{x=1}^\infty e^{tx} p_X(x) = \sum_{x=1}^\infty e^{tx} (1 - p)^{x - 1} p = \frac{p}{1 - p} \sum_{x=1}^\infty \left((1 - p) e^t\right)^x
    • Using the geometric series formula with b=(1−p)etb = (1 - p)e^t, valid when (1−p)et<1(1 - p)e^t < 1 (i.e., t<−ln⁡(1−p)t < -\ln(1 - p)) gives:     ϕX(t)=p1−p⋅(1−p)et1−(1−p)et=pet1−(1−p)et\phi_X(t) = \frac{p}{1 - p} \cdot \frac{(1 - p) e^t}{1 - (1 - p) e^t} = \frac{p e^t}{1 - (1 - p) e^t}
  • Mean and Variance:

    • The expected number of trials until the first success is the reciprocal of the success probability:     E(X)=1pE(X) = \frac{1}{p}
    • The variance is given by:     Var(X)=1−pp2\text{Var}(X) = \frac{1 - p}{p^2}
  • Application Example: Trials Until Passing the IELTS Examination:

    • Let the probability of passing the IELTS exam on any given attempt be p=0.25p = 0.25.
    • Let X∼Geom(0.25)X \sim \text{Geom}(0.25) represent the attempt number on which the candidate passes.
    • Probability of passing on the third trial:     P(X=3)=pX(3)=(1−0.25)3−1(0.25)=(0.75)2(0.25)=0.5625×0.25=0.140625=964P(X = 3) = p_X(3) = (1 - 0.25)^{3 - 1} (0.25) = (0.75)^2 (0.25) = 0.5625 \times 0.25 = 0.140625 = \frac{9}{64}
    • Expected attempt on which the candidate passes:     E(X)=1p=10.25=4E(X) = \frac{1}{p} = \frac{1}{0.25} = 4
    • Variance of the attempts:     Var(X)=1−pp2=1−0.250.252=0.750.0625=12\text{Var}(X) = \frac{1 - p}{p^2} = \frac{1 - 0.25}{0.25^2} = \frac{0.75}{0.0625} = 12
  • Visual Behavior across Different Probability Levels:

    • Geometric distributions always display a monotonic decreasing probability profile; the mode is consistently at x=1x = 1.
    • Comparison for varied pp values:
    • p=0.25  ⟹  E(X)=4p = 0.25 \implies E(X) = 4, Var(X)=12\text{Var}(X) = 12. The PMF starts at pX(1)=0.25p_X(1) = 0.25 and has a long tail.
    • p=0.50  ⟹  E(X)=2p = 0.50 \implies E(X) = 2, Var(X)=2\text{Var}(X) = 2. The PMF starts at pX(1)=0.50p_X(1) = 0.50 and halves at every step.
    • p=0.75  ⟹  E(X)≈1.33p = 0.75 \implies E(X) \approx 1.33, Var(X)≈0.44\text{Var}(X) \approx 0.44. The PMF starts at pX(1)=0.75p_X(1) = 0.75 and decays rapidly.

Geometric distribution PMF plots for p = 0.25, 0.50, and 0.75

Hypergeometric Random Variable

  • Sampling Without Replacement Framework:
    • Models a finite population of size N+MN + M containing:
    • NN items classified as Successes (e.g., Red balls).
    • MM items classified as Failures (e.g., Black balls).
    • A sample of nn items is drawn completely at random without replacement.
    • The random variable XX represents the total number of successes (Red balls) selected in the sample.

Urn model for the hypergeometric distribution with 4 red and 6 black balls

  • Derivation of the Hypergeometric PMF:

    • Total number of equally likely ways to select nn items from a population of N+MN + M items is (N+Mn)\binom{N + M}{n}.
    • The number of ways to choose xx successes from NN available successes is (Nx)\binom{N}{x}.
    • The remaining n−xn - x items in the sample must be failures, selected from the MM available failures in (Mn−x)\binom{M}{n - x} ways.
    • General Definition of the PMF for X∼Hypergeom(N,M,n)X \sim \text{Hypergeom}(N, M, n):     pX(x)=(Nx)(Mn−x)(N+Mn)for x=max⁡(0,n−M),…,min⁡(n,N)p_X(x) = \frac{\binom{N}{x} \binom{M}{n - x}}{\binom{N + M}{n}} \quad \text{for } x = \max(0, n - M), \dots, \min(n, N)
  • Detailed Urn Computation (44 Red, 66 Black, Draw 33):

    • Parameters: N=4N = 4, M=6M = 6, total items N+M=10N + M = 10, draw size n=3n = 3.
    • Support of XX: x∈{0,1,2,3}x \in \{0, 1, 2, 3\}.
    • Total outcomes in sample space: (103)=10×9×83×2×1=120\binom{10}{3} = \frac{10 \times 9 \times 8}{3 \times 2 \times 1} = 120.
    • Computation for each outcome:
    • x=0x = 0 (No red balls, 33 black balls):       pX(0)=(40)(63)(103)=1×20120=20120=16p_X(0) = \frac{\binom{4}{0} \binom{6}{3}}{\binom{10}{3}} = \frac{1 \times 20}{120} = \frac{20}{120} = \frac{1}{6}
    • x=1x = 1 (11 red ball, 22 black balls):       pX(1)=(41)(62)(103)=4×15120=60120=12p_X(1) = \frac{\binom{4}{1} \binom{6}{2}}{\binom{10}{3}} = \frac{4 \times 15}{120} = \frac{60}{120} = \frac{1}{2}
    • x=2x = 2 (22 red balls, 11 black ball):       pX(2)=(42)(61)(103)=6×6120=36120=310p_X(2) = \frac{\binom{4}{2} \binom{6}{1}}{\binom{10}{3}} = \frac{6 \times 6}{120} = \frac{36}{120} = \frac{3}{10}
    • x=3x = 3 (33 red balls, 00 black balls):       pX(3)=(43)(60)(103)=4×1120=4120=130p_X(3) = \frac{\binom{4}{3} \binom{6}{0}}{\binom{10}{3}} = \frac{4 \times 1}{120} = \frac{4}{120} = \frac{1}{30}
    • Sum check: 20120+60120+36120+4120=120120=1\frac{20}{120} + \frac{60}{120} + \frac{36}{120} + \frac{4}{120} = \frac{120}{120} = 1.
  • Moment Generating Function, Mean, and Variance:

    • The MGF ϕX(t)\phi_X(t) of the Hypergeometric distribution is mathematically not tractable in closed analytic form.
    • Expected value:     E(X)=n(NN+M)E(X) = n \left(\frac{N}{N + M}\right)
    • Defining p=NN+Mp = \frac{N}{N + M} as the initial proportion of successes in the population, the expectation matches the Binomial expectation exactly: E(X)=npE(X) = n p.
    • Variance:     Var(X)=n(NN+M)(MN+M)[1−n−1N+M−1]\text{Var}(X) = n \left(\frac{N}{N + M}\right) \left(\frac{M}{N + M}\right) \left[1 - \frac{n - 1}{N + M - 1}\right]
    • Using the algebraic identity 1−n−1N+M−1=(N+M−1)−(n−1)N+M−1=N+M−nN+M−11 - \frac{n - 1}{N + M - 1} = \frac{(N + M - 1) - (n - 1)}{N + M - 1} = \frac{N + M - n}{N + M - 1}, the variance is expressed as:       Var(X)=np(1−p)(N+M−nN+M−1)\text{Var}(X) = n p (1 - p) \left(\frac{N + M - n}{N + M - 1}\right)
    • The factor N+M−nN+M−1\frac{N + M - n}{N + M - 1} is the finite population correction factor, reflecting the reduction in variance caused by sampling without replacement compared to independent trials.