Random Variable Types

Why This Matters

Random variables are the mathematical bridge between real-world uncertainty and rigorous probability theory. When you're analyzing everything from quantum mechanics to financial markets, you're choosing which type of random variable best captures the underlying randomness. This topic connects directly to probability distributions, expected value calculations, and statistical inference—concepts that appear throughout your coursework and exams.

You're being tested on more than definitions here. Examiners want to see that you understand when to apply each distribution, what parameters define it, and how different random variables relate to one another. Can you recognize that a binomial is just repeated Bernoulli trials? Do you know why exponential and Poisson distributions are mathematically linked? Don't just memorize formulas—know what real-world scenario each random variable models and what makes it the right tool for that job.

---

Discrete vs. Continuous: The Fundamental Split

Before diving into specific distributions, you need to internalize the core distinction: discrete random variables count, continuous random variables measure. This determines everything from how we calculate probabilities to what functions describe them.

Discrete Random Variables

• Countable outcomes—these variables take on specific, separated values you could list (even if that list is infinite)
• Probability mass function (PMF) assigns exact probabilities to each value; $P(X = x)$ makes sense and can be nonzero
• Summation is used for expected value: $E[X] = \sum_x x \cdot P(X = x)$

Continuous Random Variables

• Uncountably infinite outcomes—values fill an entire interval with no gaps between possible results
• Probability density function (PDF) describes relative likelihood; $P(X = x) = 0$ for any specific value, so we integrate over intervals
• Integration replaces summation: $E[X] = \int_{-\infty}^{\infty} x \cdot f(x) \, dx$

---

Binary and Count-Based Discrete Distributions

These distributions model scenarios where you're counting successes, events, or trials. The key is identifying what's being counted and under what conditions.

Bernoulli Random Variables

• Single trial with two outcomes—the simplest random variable, taking value 1 (success) with probability $p$ or 0 (failure) with probability $1-p$
• Building block for more complex distributions; binomial, geometric, and negative binomial all derive from repeated Bernoulli trials
• Mean and variance are $E[X] = p$ and $\text{Var}(X) = p(1-p)$, both determined by the single parameter $p$

Binomial Random Variables

• Fixed number of independent trials—counts successes in $n$ Bernoulli trials, each with success probability $p$
• PMF formula: $P(X = k) = \binom{n}{k} p^k (1-p)^{n-k}$ combines counting (how many ways) with probability (how likely)
• Mean $np$ and variance $np(1-p)$ scale linearly with trial count, making this ideal for sampling with replacement

Geometric Random Variables

• Trials until first success—counts how many Bernoulli trials occur before (or including) the first success
• Memoryless property among discrete distributions; past failures don't affect future success probability
• PMF: $P(X = k) = (1-p)^{k-1}p$, with mean $1/p$—higher success probability means fewer expected trials

Poisson Random Variables

• Events in fixed intervals—models count of occurrences when events happen at a constant average rate $\lambda$
• Single parameter $\lambda$ serves as both mean and variance; useful approximation for binomial when $n$ is large and $p$ is small
• Independence assumption—events in non-overlapping intervals are independent, making this ideal for rare event modeling

Hypergeometric Random Variables

• Sampling without replacement—counts successes when drawing $n$ items from a population of $N$ containing $K$ successes
• Three parameters ($N$, $K$, $n$) capture population structure; probabilities change with each draw
• Approaches binomial when population size $N$ is much larger than sample size $n$, since replacement effects become negligible

---

Continuous Distributions for Measurement and Time

These distributions model quantities that can take any value in a range. Focus on what each distribution's shape tells you about the underlying phenomenon.

Uniform Random Variables

• Equal likelihood across an interval—every value between $a$ and $b$ is equally probable, with PDF $f(x) = \frac{1}{b-a}$
• Maximum entropy distribution when you only know the range; represents complete uncertainty within bounds
• Mean $(a+b)/2$ and variance $(b-a)^2/12$ depend only on the interval endpoints

Normal (Gaussian) Random Variables

• Bell-shaped symmetry—defined by mean $\mu$ (center) and standard deviation $\sigma$ (spread), with PDF $f(x) = \frac{1}{\sigma\sqrt{2\pi}} e^{-\frac{(x-\mu)^2}{2\sigma^2}}$
• Central Limit Theorem connection—sums of independent random variables converge to normal, explaining its ubiquity in nature
• 68-95-99.7 rule—approximately 68%, 95%, and 99.7% of values fall within 1, 2, and 3 standard deviations of the mean

Exponential Random Variables

• Waiting time until an event—models time between Poisson events, with rate parameter $\lambda$ and PDF $f(x) = \lambda e^{-\lambda x}$ for $x \geq 0$
• Memoryless property is unique among continuous distributions; $P(X > s + t \mid X > s) = P(X > t)$
• Mean $1/\lambda$ and variance $1/\lambda^2$—higher rate means shorter expected waiting time

---

Quick Reference Table

---

Self-Check Questions

1. Which two distributions share the memoryless property, and what distinguishes them from each other?

2. You're modeling the number of defective items in a batch of 20 drawn from a shipment of 100. Should you use binomial or hypergeometric, and why?

3. Compare and contrast the Poisson and exponential distributions: what real-world scenario would use both, and how are their parameters related?

4. A student claims that $P(X = 2.5) = 0.3$ for a continuous random variable. What's wrong with this statement, and how should probabilities for continuous variables be expressed?

5. If you sum 50 independent Bernoulli random variables (each with $p = 0.4$), what distribution describes the result? What distribution would the sum approximate if you instead summed 1000 such variables and standardized the result?