Skip to content

Standard Discrete and Continuous Distributions

CSIR NET statistics: binomial, Poisson, geometric, negative binomial, hypergeometric, normal, exponential, gamma, beta, Cauchy; moments, MGFs and links.

  • 21 min read
  • Updated
On this page · 13 sections
  1. Prerequisites
  2. The Hook: One Counter, Three Distributions
  3. Generating Functions: The Tool That Does the Work
  4. Discrete Distributions
  5. Continuous Distributions
  6. The Web of Relationships
  7. Memory Tricks
  8. Don’t Confuse
  9. Common Mistakes
  10. Practice
  11. Quick Revision Table
  12. Key Takeaways
  13. What’s Next

Prerequisites

The Hook: One Counter, Three Distributions

Generating Functions: The Tool That Does the Work

For a random variable \( X \):

FunctionDefinitionUse
MGF\( M_X(t) = E(e^{tX}) \) (where finite near 0)moments \( E(X^k) = M_X^{(k)}(0) \); sums of independent variables
PGF (non-negative integer \( X \))\( G_X(s) = E(s^X) \)\( P(X = k) = \frac{G^{(k)}(0)}{k!} \); \( E(X) = G'(1) \)
Characteristic function\( \varphi_X(t) = E(e^{itX}) \)always exists; determines the law
Cumulant generating function\( K_X(t) = \log M_X(t) \)\( K'(0) = \mu \), \( K''(0) = \sigma^2 \)

Two facts carry most of the exam. If \( X \) and \( Y \) are independent, \( M_{X+Y} = M_XM_Y \). And if two MGFs agree on an open interval around 0, the distributions are equal. So to identify the law of a sum, multiply MGFs and recognise the product.

For a non-negative integer variable, the variance comes fastest from the PGF: \( \operatorname{Var}(X) = G''(1) + G'(1) - G'(1)^2 \).

Discrete Distributions

Throughout, \( q = 1 - p \).

Distribution\( P(X = x) \)SupportMeanVarianceMGF
Bernoulli\( (p) \)\( p^xq^{1-x} \)\( 0, 1 \)\( p \)\( pq \)\( q + pe^t \)
Binomial\( (n, p) \)\( \binom{n}{x}p^xq^{n-x} \)\( 0, \dots, n \)\( np \)\( npq \)\( (q + pe^t)^n \)
Poisson\( (\lambda) \)\( \frac{e^{-\lambda}\lambda^x}{x!} \)\( 0, 1, 2, \dots \)\( \lambda \)\( \lambda \)\( e^{\lambda(e^t - 1)} \)
Geometric\( (p) \), failures before 1st success\( q^xp \)\( 0, 1, 2, \dots \)\( \frac{q}{p} \)\( \frac{q}{p^2} \)\( \frac{p}{1 - qe^t} \)
Negative binomial\( (r, p) \), failures before \( r \)-th success\( \binom{x + r - 1}{x}p^rq^x \)\( 0, 1, 2, \dots \)\( \frac{rq}{p} \)\( \frac{rq}{p^2} \)\( \left(\frac{p}{1 - qe^t}\right)^r \)
Hypergeometric\( (N, M, n) \)\( \frac{\binom{M}{x}\binom{N-M}{n-x}}{\binom{N}{n}} \)\( \max(0, n - N + M) \) to \( \min(n, M) \)\( n\frac{M}{N} \)\( n\frac{M}{N}\left(1 - \frac{M}{N}\right)\frac{N - n}{N - 1} \)(no simple form)
Discrete uniform on \( \{1, \dots, n\} \)\( \frac{1}{n} \)\( 1, \dots, n \)\( \frac{n + 1}{2} \)\( \frac{n^2 - 1}{12} \)\( \frac{e^t(e^{nt} - 1)}{n(e^t - 1)} \)

The MGFs of the geometric and negative binomial exist for \( qe^t < 1 \), that is \( t < -\log q \).

Points worth knowing about each

Binomial. A sum of \( n \) independent Bernoulli\( (p) \) variables. Mode: \( \lfloor (n + 1)p \rfloor \); when \( (n + 1)p \) is an integer, both \( (n + 1)p \) and \( (n + 1)p - 1 \) are modes. Variance \( npq \le \frac{n}{4} \), with equality at \( p = \frac{1}{2} \). The variance is always smaller than the mean.

Poisson. Mean equals variance, which is the quickest test for whether a count is plausibly Poisson. Skewness \( \frac{1}{\sqrt{\lambda}} \) (positive, shrinking as \( \lambda \) grows). Mode \( \lfloor \lambda \rfloor \), with \( \lambda \) and \( \lambda - 1 \) both modes when \( \lambda \) is an integer. Recurrence: \( P(X = x + 1) = \frac{\lambda}{x + 1}P(X = x) \).

Poisson as a limit of the binomial. If \( n \to \infty \) and \( p \to 0 \) with \( np = \lambda \) fixed, \( \binom{n}{x}p^xq^{n-x} \to \frac{e^{-\lambda}\lambda^x}{x!} \). Proof by MGFs: \( (1 + \frac{\lambda(e^t - 1)}{n})^n \to e^{\lambda(e^t - 1)} \).

Geometric. The only discrete distribution on \( \{0, 1, 2, \dots\} \) with the memoryless property: \( P(X \ge m + n \mid X \ge m) = P(X \ge n) \), because \( P(X \ge k) = q^k \).

Negative binomial. A sum of \( r \) independent geometric variables. Its variance exceeds its mean (\( \frac{rq}{p^2} > \frac{rq}{p} \)), so it is the standard model for over-dispersed counts.

Hypergeometric. Sampling \( n \) items without replacement from \( N \), of which \( M \) are “successes”. The mean is the binomial mean \( n\frac{M}{N} \); the variance is the binomial variance times the finite population correction \( \frac{N - n}{N - 1} < 1 \). As \( N \to \infty \) with \( \frac{M}{N} \to p \), it tends to Binomial\( (n, p) \).

Multinomial. \( n \) trials, each landing in category \( i \) with probability \( p_i \): \( P(X_1 = x_1, \dots, X_k = x_k) = \frac{n!}{x_1!\cdots x_k!}p_1^{x_1}\cdots p_k^{x_k} \). Each \( X_i \sim \) Binomial\( (n, p_i) \), and \( \operatorname{Cov}(X_i, X_j) = -np_ip_j \) for \( i \ne j \): the counts are negatively correlated because they share a fixed total.

Continuous Distributions

DistributionDensitySupportMeanVarianceMGF
Uniform\( (a, b) \)\( \frac{1}{b - a} \)\( (a, b) \)\( \frac{a + b}{2} \)\( \frac{(b - a)^2}{12} \)\( \frac{e^{tb} - e^{ta}}{t(b - a)} \)
Normal\( (\mu, \sigma^2) \)\( \frac{1}{\sigma\sqrt{2\pi}}e^{-\frac{(x - \mu)^2}{2\sigma^2}} \)\( \mathbb{R} \)\( \mu \)\( \sigma^2 \)\( e^{\mu t + \frac{\sigma^2t^2}{2}} \)
Exponential, rate \( \lambda \)\( \lambda e^{-\lambda x} \)\( x > 0 \)\( \frac{1}{\lambda} \)\( \frac{1}{\lambda^2} \)\( \frac{\lambda}{\lambda - t} \), \( t < \lambda \)
Gamma, shape \( \alpha \), rate \( \lambda \)\( \frac{\lambda^\alpha}{\Gamma(\alpha)}x^{\alpha - 1}e^{-\lambda x} \)\( x > 0 \)\( \frac{\alpha}{\lambda} \)\( \frac{\alpha}{\lambda^2} \)\( \left(\frac{\lambda}{\lambda - t}\right)^\alpha \)
Beta\( (a, b) \)\( \frac{x^{a-1}(1 - x)^{b-1}}{B(a, b)} \)\( (0, 1) \)\( \frac{a}{a + b} \)\( \frac{ab}{(a + b)^2(a + b + 1)} \)(no simple form)
Cauchy\( (0, 1) \)\( \frac{1}{\pi(1 + x^2)} \)\( \mathbb{R} \)does not existdoes not existdoes not exist; \( \varphi(t) = e^{-\lvert t \rvert} \)
Laplace\( (\mu, b) \)\( \frac{1}{2b}e^{-\lvert x - \mu \rvert/b} \)\( \mathbb{R} \)\( \mu \)\( 2b^2 \)\( \frac{e^{\mu t}}{1 - b^2t^2} \), \( \lvert t \rvert < \frac{1}{b} \)
Lognormal: \( \log X \sim N(\mu, \sigma^2) \)\( \frac{1}{x\sigma\sqrt{2\pi}}e^{-\frac{(\log x - \mu)^2}{2\sigma^2}} \)\( x > 0 \)\( e^{\mu + \sigma^2/2} \)\( (e^{\sigma^2} - 1)e^{2\mu + \sigma^2} \)does not exist for \( t > 0 \)

The normal distribution

  • Symmetric about \( \mu \); mean = median = mode \( = \mu \); points of inflection at \( \mu \pm \sigma \).
  • \( Z = \frac{X - \mu}{\sigma} \sim N(0, 1) \). Central moments: odd ones vanish, \( E(X - \mu)^{2k} = (2k - 1)(2k - 3)\cdots 3 \cdot 1 \cdot \sigma^{2k} \). So \( \mu_4 = 3\sigma^4 \), \( \beta_2 = 3 \), \( \beta_1 = 0 \).
  • Mean absolute deviation about the mean: \( \sigma\sqrt{\frac{2}{\pi}} \approx 0.798\sigma \).
  • Areas: \( P(\lvert Z \rvert < 1) \approx 0.683 \), \( P(\lvert Z \rvert < 1.96) = 0.95 \), \( P(\lvert Z \rvert < 2) \approx 0.954 \), \( P(\lvert Z \rvert < 2.576) = 0.99 \), \( P(\lvert Z \rvert < 3) \approx 0.997 \).
  • Any linear combination of independent normals is normal: \( \sum a_iX_i \sim N(\sum a_i\mu_i, \sum a_i^2\sigma_i^2) \).
  • If \( X, Y \) are i.i.d. \( N(\mu, \sigma^2) \), then \( X + Y \) and \( X - Y \) are independent. (They are jointly normal with covariance \( \sigma^2 - \sigma^2 = 0 \).)

Exponential and gamma

  • Memoryless: \( P(X > s + t \mid X > s) = P(X > t) = e^{-\lambda t} \). Among continuous distributions on \( (0, \infty) \), only the exponential has this property.
  • Median \( \frac{\log 2}{\lambda} \), less than the mean \( \frac{1}{\lambda} \) (right skew: \( \gamma_1 = \frac{\mu_3}{\sigma^3} = 2 \), so \( \beta_1 = 4 \); \( \beta_2 = 9 \)).
  • The hazard rate \( \frac{f(x)}{1 - F(x)} = \lambda \) is constant.
  • Gamma mode \( \frac{\alpha - 1}{\lambda} \) for \( \alpha \ge 1 \). Gamma\( (1, \lambda) \) is Exponential\( (\lambda) \).
  • \( \Gamma(\alpha + 1) = \alpha\Gamma(\alpha) \), \( \Gamma(n) = (n - 1)! \), \( \Gamma\left(\frac{1}{2}\right) = \sqrt{\pi} \).

Beta

  • Beta\( (1, 1) \) is Uniform\( (0, 1) \). Mode \( \frac{a - 1}{a + b - 2} \) for \( a, b > 1 \). If \( X \sim \) Beta\( (a, b) \) then \( 1 - X \sim \) Beta\( (b, a) \).
  • Moments: \( E(X^k) = \frac{B(a + k, b)}{B(a, b)} \).

Cauchy

  • Heavy tails: \( E\lvert X \rvert = \infty \), so no mean, no variance, no MGF. Median 0, quartiles \( \pm 1 \).
  • The ratio of two independent \( N(0, 1) \) variables is Cauchy; so is \( \tan U \) for \( U \sim \) Uniform\( (-\frac{\pi}{2}, \frac{\pi}{2}) \).
  • If \( X_1, \dots, X_n \) are i.i.d. standard Cauchy, \( \bar{X} \) is again standard Cauchy (from \( \varphi_{\bar{X}}(t) = \varphi(t/n)^n = e^{-\lvert t \rvert} \)). Averaging does not help, and the law of large numbers fails.

The Web of Relationships

graph TD
    B["Bernoulli(p)"] -->|"sum of n i.i.d."| Bin["Binomial(n, p)"]
    Bin -->|"n large, np = λ fixed"| P["Poisson(λ)"]
    H["Hypergeometric(N, M, n)"] -->|"N large, M/N = p"| Bin
    G["Geometric(p)"] -->|"sum of r i.i.d."| NB["Negative binomial(r, p)"]
    E["Exponential(λ)"] -->|"sum of n i.i.d."| Ga["Gamma(n, λ)"]
    U["Uniform(0, 1)"] -->|"-log U"| E
    Ga -->|"shape n/2, rate 1/2"| C["Chi-square(n)"]
    N["Normal(0, 1)"] -->|"sum of n squares"| C
    N -->|"ratio of two independent"| Ca["Cauchy(0, 1)"]
    Ga -->|"X/(X+Y), same rate"| Be["Beta(α, β)"]
    Bin -->|"n large: CLT"| N
    P -->|"λ large: CLT"| N

Reproductive (additive) properties

For independent variables:

If \( X_i \sim \)then \( \sum X_i \sim \)
Binomial\( (n_i, p) \), same \( p \)Binomial\( (\sum n_i, p) \)
Poisson\( (\lambda_i) \)Poisson\( (\sum \lambda_i) \)
Negative binomial\( (r_i, p) \), same \( p \)Negative binomial\( (\sum r_i, p) \)
Gamma\( (\alpha_i, \lambda) \), same rateGamma\( (\sum \alpha_i, \lambda) \)
\( \chi^2_{n_i} \)\( \chi^2_{\sum n_i} \)
\( N(\mu_i, \sigma_i^2) \)\( N(\sum \mu_i, \sum \sigma_i^2) \)
Cauchy, location \( a_i \), scale \( b_i \)Cauchy, location \( \sum a_i \), scale \( \sum b_i \)

Uniform, exponential (as exponential), binomial with different \( p \): not closed under sums. The sum of two independent Uniform\( (0, 1) \) is triangular on \( (0, 2) \).

Transformations and conditionals to know cold

  1. Probability integral transform. If \( F \) is continuous and \( X \sim F \), then \( F(X) \sim \) Uniform\( (0, 1) \). Conversely \( F^{-1}(U) \sim F \). So \( -\frac{1}{\lambda}\log U \sim \) Exponential\( (\lambda) \), and \( -2\log U \sim \chi^2_2 \).
  2. Minimum of exponentials. If \( X_i \sim \) Exponential\( (\lambda_i) \) independently, \( \min X_i \sim \) Exponential\( (\sum \lambda_i) \), and \( P(X_1 \text{ is the smallest}) = \frac{\lambda_1}{\sum \lambda_i} \).
  3. Poisson splitting. If \( X \sim \) Poisson\( (\lambda_1) \), \( Y \sim \) Poisson\( (\lambda_2) \) independent, then given \( X + Y = n \), \( X \sim \) Binomial\( \left(n, \frac{\lambda_1}{\lambda_1 + \lambda_2}\right) \).
  4. Poisson thinning. If \( N \sim \) Poisson\( (\lambda) \) and each of the \( N \) events is kept independently with probability \( p \), the kept count is Poisson\( (\lambda p) \), independent of the discarded count.
  5. Gamma to beta. If \( X \sim \) Gamma\( (\alpha, \lambda) \), \( Y \sim \) Gamma\( (\beta, \lambda) \) independent, then \( \frac{X}{X + Y} \sim \) Beta\( (\alpha, \beta) \), independent of \( X + Y \).
  6. Gamma and Poisson. In a Poisson process of rate \( \lambda \), the time \( T_k \) of the \( k \)-th event is Gamma\( (k, \lambda) \), and \( P(T_k > t) = P(\text{Poisson}(\lambda t) \le k - 1) \).
  7. Binomial and beta. \( P(\text{Binomial}(n, p) \ge k) = P(\text{Beta}(k, n - k + 1) \le p) \). (Both are the probability that the \( k \)-th smallest of \( n \) uniforms is below \( p \).)

Memory Tricks

  • “Poisson: mean is variance.” Binomial: variance below mean. Negative binomial: variance above mean. Under, equal, over.
  • “Exponential forgets, gamma counts.” Exponential is the wait for one event; gamma is the wait for \( k \); Poisson is how many events fit in a window.
  • MGF shapes. \( (q + pe^t)^n \) binomial; \( e^{\lambda(e^t - 1)} \) Poisson; \( (1 - t/\lambda)^{-\alpha} \) gamma; \( e^{\mu t + \sigma^2t^2/2} \) normal. Spot the shape, read off the parameters.
  • Finite population correction \( \frac{N - n}{N - 1} \). Without replacement, the variance shrinks; with \( n = N \) it is zero (you have counted everything).
  • Uniform variance = (length) squared over 12. Continuous \( \frac{(b - a)^2}{12} \); discrete on \( 1, \dots, n \) is \( \frac{n^2 - 1}{12} \).
  • Cauchy breaks everything. No mean, no LLN, sample mean as bad as one observation. It is the default counter-example in Part C.

Don’t Confuse

ThisIs not thisThe difference
geometric: failures before success (mean \( q/p \))geometric: trials to success (mean \( 1/p \))shift by 1; same variance \( q/p^2 \)
exponential with rate \( \lambda \) (mean \( 1/\lambda \))exponential with mean \( \theta \)read the density given
binomial variance \( npq \)hypergeometric varianceextra factor \( \frac{N - n}{N - 1} \)
uncorrelated normals that are jointly normalany two uncorrelated normalsonly joint normality makes zero covariance mean independence
Poisson\( (\lambda_1) + \) Poisson\( (\lambda_2) \) is PoissonPoisson\( (\lambda_1) - \) Poisson\( (\lambda_2) \)a difference takes negative values; it is not Poisson
Cauchy: median existsCauchy: mean exists\( E\lvert X \rvert = \infty \)
lognormal median \( e^\mu \)lognormal meanthe mean is \( e^{\mu + \sigma^2/2} \)

Common Mistakes

Practice

All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.

Level 1: Foundation

  1. State the mean and variance of Binomial\( (10, 0.3) \).
  2. Which continuous distribution on \( (0, \infty) \) is memoryless?
  3. If \( X \sim N(2, 9) \), what is the distribution of \( \frac{X - 2}{3} \)?
Answers
  1. Mean \( np = 3 \), variance \( npq = 3 \times 0.7 = 2.1 \).
  2. The exponential distribution.
  3. \( N(0, 1) \). (Note that 9 is the variance, so \( \sigma = 3 \).)

Level 2: Part B Style

Practice Q1. \( X \) is Poisson with \( P(X = 1) = P(X = 2) \). Then \( P(X = 0) \) is:

(a) \( e^{-1} \) (b) \( e^{-2} \) (c) \( 2e^{-2} \) (d) \( \frac{1}{2} \)

Answer: (b)

\( \lambda e^{-\lambda} = \frac{\lambda^2}{2}e^{-\lambda} \) gives \( \lambda = 2 \). So \( P(X = 0) = e^{-2} \approx 0.135 \). Option (c) is \( P(X = 1) \).

Practice Q2. A binomial variable has mean 4 and variance 3. The number of trials \( n \) is:

(a) 12 (b) 16 (c) 18 (d) 20

Answer: (b)

\( q = \frac{npq}{np} = \frac{3}{4} \), so \( p = \frac{1}{4} \) and \( n = \frac{4}{1/4} = 16 \).

Practice Q3. \( X \) is exponential with mean 2. Then \( P(X > 5 \mid X > 3) \) equals:

(a) \( e^{-1} \) (b) \( e^{-2} \) (c) \( e^{-5/2} \) (d) \( \frac{2}{5} \)

Answer: (a)

Rate \( \lambda = \frac{1}{2} \). By memorylessness, \( P(X > 5 \mid X > 3) = P(X > 2) = e^{-2/2} = e^{-1} \approx 0.368 \). Option (b) uses rate 2 instead of mean 2.

Practice Q4. \( U \sim \) Uniform\( (0, 1) \). The distribution of \( -2\log U \) is:

(a) exponential with mean \( \frac{1}{2} \) (b) \( \chi^2 \) with 2 degrees of freedom (c) Uniform\( (0, 2) \) (d) gamma with shape 2 and rate 1

Answer: (b)

\( P(-2\log U > y) = P(U < e^{-y/2}) = e^{-y/2} \): exponential with mean 2, which is Gamma\( (1, \frac{1}{2}) = \chi^2_2 \). Option (a) has the mean inverted.

Level 3: Part C Style

Practice Q5. Let \( X_1, \dots, X_n \) (\( n \ge 2 \)) be i.i.d. exponential with rate \( \lambda \). Which statements are true?

(a) \( \min(X_1, \dots, X_n) \) is exponential with rate \( n\lambda \). (b) \( X_1 + \dots + X_n \) has a gamma distribution with shape \( n \) and rate \( \lambda \). (c) \( 2\lambda(X_1 + \dots + X_n) \sim \chi^2_{2n} \). (d) \( \max(X_1, \dots, X_n) \) is exponential with rate \( \frac{\lambda}{n} \).

Answer: (a), (b) and (c)
  • (a) True. \( P(\min > x) = (e^{-\lambda x})^n = e^{-n\lambda x} \).
  • (b) True. MGF \( \left(\frac{\lambda}{\lambda - t}\right)^n \).
  • (c) True. Scaling a Gamma\( (n, \lambda) \) by \( 2\lambda \) gives Gamma\( (n, \frac{1}{2}) = \chi^2_{2n} \). This is the pivot for exponential confidence intervals.
  • (d) False. \( P(\max \le x) = (1 - e^{-\lambda x})^n \), not of the form \( 1 - e^{-cx} \). Its mean is \( \frac{1}{\lambda}\sum_{k=1}^n \frac{1}{k} \), not \( \frac{n}{\lambda} \).

Practice Q6. \( X \sim \) Poisson(1) and \( Y \sim \) Poisson(2) are independent. Which statements are true?

(a) \( X + Y \sim \) Poisson(3). (b) \( P(X = 1 \mid X + Y = 3) = \frac{4}{9} \). (c) \( X - Y \) is Poisson with mean \( -1 \). (d) \( \operatorname{Var}(X - Y) = 3 \).

Answer: (a), (b) and (d)
  • (a) True. Additive property.
  • (b) True. Given \( X + Y = 3 \), \( X \sim \) Binomial\( (3, \frac{1}{3}) \): \( P(X = 1) = 3 \cdot \frac{1}{3} \cdot \frac{4}{9} = \frac{4}{9} \).
  • (c) False. A Poisson mean cannot be negative; \( X - Y \) takes negative values.
  • (d) True. Independent variances add: \( 1 + 2 = 3 \).

Practice Q7. An urn has 10 balls, 4 red. Five balls are drawn. Let \( X \) be the number of red balls drawn without replacement and \( Y \) the number drawn with replacement. Which statements are true?

(a) \( E(X) = E(Y) = 2 \). (b) \( \operatorname{Var}(X) = \frac{2}{3} \). (c) \( P(X = 0) = \frac{1}{42} \). (d) \( \operatorname{Var}(X) > \operatorname{Var}(Y) \).

Answer: (a), (b) and (c)
  • (a) True. Both means are \( 5 \cdot \frac{4}{10} = 2 \).
  • (b) True. \( \operatorname{Var}(Y) = 5 \cdot 0.4 \cdot 0.6 = 1.2 \); \( \operatorname{Var}(X) = 1.2 \cdot \frac{10 - 5}{10 - 1} = 1.2 \cdot \frac{5}{9} = \frac{2}{3} \).
  • (c) True. \( \frac{\binom{6}{5}}{\binom{10}{5}} = \frac{6}{252} = \frac{1}{42} \).
  • (d) False. The finite population correction makes the hypergeometric variance smaller.

Practice Q8. \( Z \sim N(0, 1) \). Which statements are true?

(a) \( E\lvert Z \rvert = \sqrt{2/\pi} \). (b) \( E(Z^4) = 3 \). (c) \( Z^2 \) has a gamma distribution with shape \( \frac{1}{2} \) and rate \( \frac{1}{2} \). (d) \( P(\lvert Z \rvert > 2) > 0.1 \).

Answer: (a), (b) and (c)
  • (a) True. \( 2\int_0^\infty z\frac{e^{-z^2/2}}{\sqrt{2\pi}}\,dz = \frac{2}{\sqrt{2\pi}} = \sqrt{\frac{2}{\pi}} \approx 0.798 \).
  • (b) True. \( \mu_4 = 3\sigma^4 \) with \( \sigma = 1 \).
  • (c) True. \( Z^2 \sim \chi^2_1 = \) Gamma\( (\frac{1}{2}, \frac{1}{2}) \).
  • (d) False. \( P(\lvert Z \rvert > 2) \approx 0.0455 \).

Quick Revision Table

DistributionMeanVarianceKey property
Binomial\( (n, p) \)\( np \)\( npq \)sum of Bernoullis; mode \( \lfloor (n+1)p \rfloor \)
Poisson\( (\lambda) \)\( \lambda \)\( \lambda \)binomial limit; additive; splitting gives binomial
Geometric (failures)\( q/p \)\( q/p^2 \)memoryless (discrete)
Negative binomial\( rq/p \)\( rq/p^2 \)sum of geometrics; over-dispersed
Hypergeometric\( nM/N \)binomial \( \times \frac{N-n}{N-1} \)without replacement
Uniform\( (a, b) \)\( \frac{a+b}{2} \)\( \frac{(b-a)^2}{12} \)\( F(X) \sim U(0, 1) \)
Normal\( \mu \)\( \sigma^2 \)\( \beta_2 = 3 \); MAD \( \sigma\sqrt{2/\pi} \)
Exponential\( (\lambda) \)\( 1/\lambda \)\( 1/\lambda^2 \)memoryless; min of exponentials
Gamma\( (\alpha, \lambda) \)\( \alpha/\lambda \)\( \alpha/\lambda^2 \)sum of exponentials; \( \chi^2_n = \) Gamma\( (\frac{n}{2}, \frac{1}{2}) \)
Beta\( (a, b) \)\( \frac{a}{a+b} \)\( \frac{ab}{(a+b)^2(a+b+1)} \)\( \frac{X}{X+Y} \) of gammas; uniform order statistics
Cauchynonenoneratio of normals; \( \bar{X} \) is Cauchy

Key Takeaways

What’s Next