Exponential Distribution: Formula, Applications, and Worked Examples
The exponential distribution explained — PDF, CDF, mean, variance, the memoryless property, and how to use it for time-between-events problems (call center waiting, equipment failure, queue arrivals) with three worked examples.
What You'll Learn
- ✓Compute probabilities, percentiles, and expected values for exponentially distributed variables.
- ✓Recognize problem patterns (waiting times, lifetimes, gap between Poisson events) that call for the exponential distribution.
- ✓Understand the memoryless property and why it limits the realism of pure exponential models.
1. Direct Answer: When to Reach for the Exponential Distribution
The exponential distribution models the time between events in a Poisson process — the waiting time until the next customer arrives, the next radioactive decay, the next equipment failure, the next phone call. Its PDF is f(x) = λ exp(-λx) for x ≥ 0, where λ (lambda) is the rate parameter (events per unit time). Mean = 1/λ. Variance = 1/λ². CDF is P(X ≤ x) = 1 - exp(-λx). It is the continuous counterpart of the geometric distribution: where the geometric counts trials until the first success in discrete trials, the exponential measures continuous time until the first event. The single most important property — and limitation — is memorylessness: P(X > s + t | X > s) = P(X > t). The remaining waiting time has the same distribution regardless of how long you have already waited.
Key Points
- •PDF: f(x) = λ exp(-λx) for x ≥ 0; CDF: F(x) = 1 - exp(-λx).
- •Mean = 1/λ, Variance = 1/λ², Standard deviation = 1/λ.
- •Models time between events in a Poisson process.
- •Memoryless: the past does not affect future waiting time.
2. Worked Example 1: Call Center Waiting Time
A call center receives calls at a rate of λ = 12 per hour (so on average one call every 5 minutes). What is the probability the next call arrives within 3 minutes? Convert to consistent units: 3 minutes = 0.05 hours. P(X ≤ 0.05) = 1 - exp(-12 × 0.05) = 1 - exp(-0.6) = 1 - 0.5488 = 0.4512. About a 45% chance. What is the probability the next call takes more than 10 minutes? P(X > 10/60) = exp(-12 × 1/6) = exp(-2) = 0.1353. About 14%. The expected waiting time is 1/λ = 1/12 hour = 5 minutes, but because the distribution is right-skewed the median is shorter: median = ln(2)/λ ≈ 0.693/12 hour ≈ 3.47 minutes. The mean exceeds the median in any right-skewed distribution.
Key Points
- •Keep λ and time in consistent units before computing.
- •P(X > t) = exp(-λt) is the survival function.
- •For exponential: median = ln(2)/λ ≈ 0.693 × mean, less than the mean.
3. Worked Example 2: Light Bulb Lifetime
A light bulb has exponentially distributed lifetime with mean 1,000 hours, so λ = 1/1000 = 0.001. What is the probability it lasts more than 1,500 hours? P(X > 1,500) = exp(-0.001 × 1,500) = exp(-1.5) = 0.2231. About 22%. What is the 90th-percentile lifetime (the value below which 90% of bulbs fail)? Solve 0.90 = 1 - exp(-0.001x) → exp(-0.001x) = 0.10 → -0.001x = ln(0.10) = -2.303 → x = 2,303 hours. That is the planning horizon for a 90% replacement schedule. Now apply memorylessness: given the bulb has already lasted 500 hours, the probability it lasts another 1,500 hours is exp(-0.001 × 1,500) = 0.2231 — the same 22%. The past 500 hours of operation do not affect the remaining distribution. This is exactly why pure exponential is unrealistic for components that wear out: real light bulbs become MORE likely to fail as they age, which the Weibull distribution captures better.
Key Points
- •Exponential is the natural model for "constant hazard rate" lifetimes.
- •Memorylessness predicts no aging — only realistic for pure-random-failure processes.
- •Use Weibull when the failure rate changes with age.
4. Worked Example 3: Connection to Poisson
A web server receives requests according to a Poisson process with λ = 30 per second. Two questions, one Poisson, one exponential. (a) In a one-second window, what is the probability of receiving exactly 25 requests? Use Poisson: P(N=25) = exp(-30) × 30^25 / 25! ≈ 0.0511. (b) Starting now, what is the probability the next request arrives within 0.02 seconds? Use exponential with rate 30 per second: P(X ≤ 0.02) = 1 - exp(-30 × 0.02) = 1 - exp(-0.6) = 0.4512. Same λ, different question, different distribution. Counting events in fixed time → Poisson. Waiting for the next event in continuous time → exponential. These are two sides of the same coin: if events follow a Poisson process with rate λ, the inter-event times are independent and exponentially distributed with rate λ.
Key Points
- •Poisson counts events per unit time; exponential measures time between events.
- •A Poisson process has both descriptions simultaneously.
- •Same rate parameter λ appears in both distributions for the same process.
5. The Memoryless Property — and Why It Matters
Memorylessness: P(X > s + t | X > s) = P(X > t). The probability of waiting an additional t units, given you have already waited s units, equals the unconditional probability of waiting t units from a fresh start. The exponential distribution is the ONLY continuous distribution with this property (geometric is the only discrete one). This is mathematically elegant and computationally convenient — but it implies no aging, no wear-out, no learning, no buildup. Real-world processes that fit: radioactive decay, well-behaved Poisson event-gap processes, certain genuinely random failures. Real-world processes that DO NOT fit: mechanical wear (bearings, brakes), biological survival (most diseases), human reaction times, customer patience (people give up after a threshold). For these, lognormal, Weibull, or gamma distributions are more realistic.
Key Points
- •Exponential is the unique continuous memoryless distribution.
- •Implies constant hazard rate — no aging or improvement.
- •For wear-out or learning processes, use Weibull, lognormal, or gamma instead.
6. Relationship to the Gamma Distribution
The sum of k independent exponential(λ) random variables follows a gamma(k, λ) distribution. This makes the gamma the natural distribution for "time to the k-th event" in a Poisson process. With k=1, gamma reduces to exponential. With integer k, gamma is sometimes called the Erlang distribution and is widely used in queueing theory and telecom engineering. Mean of gamma(k, λ) is k/λ. This relationship lets you derive the exponential from a Poisson process and the gamma by counting waiting times for multiple events.
Key Points
- •Sum of k exponentials with rate λ → Gamma(k, λ).
- •Gamma(1, λ) = exponential.
- •Used to model "time until k-th event" in a Poisson process.
7. Using StatsIQ for Exponential Problems
Snap a photo of any exponential distribution problem — call center waiting times, equipment failure questions, queueing analyses — and StatsIQ computes the requested probability, percentile, or expected value step by step with the exponential formulas labeled. The app also flags when the memorylessness assumption looks implausible for your scenario and suggests Weibull or gamma alternatives.
Key Points
- •Step-by-step solver for waiting-time problems.
- •Assumption-check warnings for non-memoryless processes.
- •Suggests Weibull/gamma alternatives when appropriate.
Key Takeaways
- ★PDF: f(x) = λ exp(-λx), x ≥ 0. CDF: F(x) = 1 - exp(-λx).
- ★Mean = 1/λ, variance = 1/λ², standard deviation = 1/λ.
- ★P(X > t) = exp(-λt) (survival function).
- ★Memoryless: P(X > s + t | X > s) = P(X > t).
- ★Inter-event times in a Poisson process are exponentially distributed.
- ★Sum of k independent exponentials = Gamma(k, λ).
Practice Questions
1. Buses arrive according to a Poisson process with rate 4 per hour. What is the probability you wait more than 20 minutes?
2. A component has exponential lifetime with mean 500 hours. What is the median lifetime?
3. You receive emails at rate 6/hour. What is the probability of receiving exactly 2 emails in 30 minutes?
FAQs
Common questions about this topic
When the process has memory — when aging, wear-out, learning, or fatigue make the future depend on the past. Mechanical components wearing out, biological survival with disease progression, customer patience that runs out at thresholds — all violate memorylessness. The Weibull distribution can model both increasing and decreasing failure rates and is the standard replacement. Lognormal works well for processes whose log is normally distributed. Gamma generalizes the exponential and works for sums of independent exponential waits.
The maximum likelihood estimator is simply 1 divided by the sample mean. If your sample mean waiting time is 7.5 minutes, the MLE for λ is 1/7.5 per minute. The MLE is unbiased only asymptotically; for small samples it is biased upward. Confirm the exponential fit with a Q-Q plot or a Kolmogorov-Smirnov test before relying on probability calculations.
They describe one process from three angles. A Poisson process with rate λ generates events whose count in any time interval is Poisson distributed, whose gap between successive events is exponential(λ), and whose time-to-the-k-th-event is gamma(k, λ). The three distributions are tied together by the same rate parameter and the same underlying process.
Because the PDF decays from a maximum at x=0 toward zero as x increases, with no upper bound. The bulk of probability is concentrated near zero, but the tail extends indefinitely. Mean exceeds median by a fixed ratio of 1/ln(2) ≈ 1.443. There is no "symmetric" exponential — symmetry would require equal probability on both sides of the median, which the decay shape forbids.
Yes — the sum is gamma-distributed, not exponential. Specifically, X1 + X2 where each is exponential(λ) gives X1 + X2 ~ Gamma(2, λ) with mean 2/λ and variance 2/λ². This is why "time until 2 events" is gamma, not exponential. Sums of exponentials with DIFFERENT rates produce a hypoexponential distribution, which has no simple closed-form PDF but is computationally tractable.
Snap a photo of any exponential distribution problem and StatsIQ identifies the rate parameter, computes the required probability, percentile, or moment step by step, and flags assumption issues (such as evidence the data has memory rather than being memoryless). For exam prep, the app generates practice problems calibrated to your difficulty level. This content is for educational purposes only.