Other Common Distributions (Recognition Only)

This page collects distributions that appear occasionally in lectures and prelim problems, usually as contrasts, counterexamples, or modeling alternatives. You are expected to recognize them, not derive them.


Cauchy Distribution

Definition:
\(X = \frac{Z_1}{Z_2}, \quad Z_1,Z_2 \sim \mathcal{N}(0,1)\)

Key properties:

  • No mean, no variance
  • Heavy tails
  • Stable under addition
  • No LLN, no CLT

Why it appears:
Counterexample to moment-based arguments.


Laplace (Double Exponential) Distribution

(Also called the L1 version of the Normal)

PDF:
\(f(x)=\frac{1}{2b}e^{-|x-\mu|/b}\)

Key properties:

  • Symmetric
  • Sharper peak than Normal
  • Heavier tails than Normal
  • MGF exists

Why it appears:

  • LASSO and $ L^1 $ regularization
  • Robust alternatives to Normal

Logistic Distribution

CDF:
\(F(x)=\frac{1}{1+e^{-(x-\mu)/s}}\)

Key properties:

  • Symmetric
  • Similar to Normal
  • Slightly heavier tails

Why it appears:

  • Logistic regression
  • Binary response models

Weibull Distribution

PDF:
\(f(x)=\frac{k}{\lambda} \left(\frac{x}{\lambda}\right)^{k-1} e^{-(x/\lambda)^k}\)

Key properties:

  • Flexible hazard rate
  • Includes Exponential as special case ($k=1$)

Why it appears:

  • Survival analysis
  • Reliability modeling

Point Mass / Degenerate Distribution

Definition:
\(\mathbb{P}(X=c)=1\)

Key properties:

  • Zero variance
  • Used in mixtures
  • Appears in conditioning arguments

Why it appears:

  • Conditional distributions
  • Mixture models
  • Edge cases in proofs

Atomic + Continuous Mixtures

Example:
\(\mathbb{P}(X=0)=p, \quad X \mid X>0 \sim F\)

Why it appears:

  • Zero-inflated models
  • Spike-and-slab priors
  • Model selection

Summary Table

Distribution Core Role
Cauchy Counterexample
Laplace Robust alternative
Logistic Binary models
Weibull Survival
Point mass Conditioning
Mixtures Sparsity

Exam Takeaways

  • Recognize, don’t derive
  • Used to break assumptions
  • Often defined in the problem
  • Know what fails (moments, CLT, etc.)

Degenerate (Dirac) Distribution — $ \delta_c $

A degenerate distribution places all probability mass at a single point.


Definition

\[\mathbb{P}(X = c) = 1\]

Equivalently, the probability measure is \(\delta_c(A) = \begin{cases} 1, & c \in A \\ 0, & c \notin A \end{cases}\)


Properties

  • Mean: $ c $
  • Variance: $ 0 $
  • Support: single point
  • All moments exist

Role in Probability

Used to describe:

  • Limits of random variables
  • Conditioning on events of probability 1
  • Mixture models (spike-and-slab)
  • Weak convergence to constants

Exam Takeaways

  • “Degenerate” = point mass
  • Limit to a constant ⇒ Dirac measure
  • Variance zero does not imply randomness

Comments