The exponential distribution calculator above works with the continuous distribution that models waiting time when events arrive at a constant average rate and independently of one another. Give it a rate or a mean and it returns the probability density at a point, the cumulative probability below it, the survival probability above it, the probability of landing between two values, any percentile you ask for, and a direct demonstration of the memoryless property that makes this distribution behave so unlike intuition.
Arb Digital publishes this page because the exponential distribution is the default model for time-to-next-event in queueing, reliability and traffic analysis, and because it is routinely applied where its central assumption does not hold. A constant rate means the chance of an event in the next second never changes, no matter how long you have waited or how old the component is. That is true of radioactive decay and roughly true of arrivals at a busy call centre. It is emphatically not true of mechanical wear, human patience, or anything with a lifecycle.
What This Exponential Distribution Calculator Does
It evaluates the single-parameter exponential distribution in both directions. Forward, it takes a value and returns probabilities. Backward, it takes a probability and returns the value — the percentile, or equivalently the inverse CDF. It also reports the mean, median, standard deviation and variance, which for this distribution are tied together in an unusually rigid way, and it runs the memoryless comparison automatically so you can see the property rather than take it on trust.
The parameterisation toggle exists because half the literature uses the rate λ and the other half uses the scale 1/λ, and the two are reciprocals. Entering a mean of 2 where the tool expects a rate of 2 does not give a slightly wrong answer, it gives a distribution four times narrower. The toggle removes that failure mode entirely.
The exponential is the continuous partner of two distributions we cover separately. Our Poisson distribution calculator counts how many events occur in a fixed window; this page measures how long you wait for one. Our geometric distribution calculator answers the same waiting question in discrete trials rather than continuous time. All three describe the same underlying process from different angles.
How to Use It
- Pick the parameterisation. Choose rate if your source says "3 arrivals per hour" and mean if it says "one arrival every 20 minutes".
- Enter the parameter. Keep the time units consistent throughout: a rate per hour needs values entered in hours.
- Set a and b. The headline cumulative probability uses a alone. The interval result uses both, and b should be larger than a.
- Choose a percentile. The 90th percentile answers questions like "how long do I need to allow to cover nine out of ten waits".
- Read the memoryless panel. It compares the chance of waiting a further a units after some time has elapsed against the chance of waiting a units from a standing start.
The Formulas and How They Are Calculated
The density is f(x) = λe−λx for x ≥ 0 and zero otherwise. Integrating it gives the cumulative distribution function F(x) = 1 − e−λx, and the survival function is simply S(x) = e−λx. Because the CDF inverts in closed form, the percentile is exact rather than iterative: xp = −ln(1 − p)/λ.
The moments are tightly linked. The mean is 1/λ, the standard deviation is also 1/λ, and the variance is 1/λ2. Mean and standard deviation being equal is a strong, checkable signature: if your data has a mean of 10 and a standard deviation of 3, it is not exponential, and no choice of λ will make it so. The median is ln(2)/λ, roughly 0.693 of the mean, which means over 63 percent of observations fall below the mean — the distribution is right-skewed and the mean is pulled up by a long tail. The NIST/SEMATECH e-Handbook page on the exponential distribution gives the same functions in the location-scale form used in reliability work.
The hazard function — the instantaneous failure rate given survival so far — is f(x)/S(x), which for the exponential is exactly λ, a constant. That single fact is the whole distribution. Constant hazard is what memorylessness means mechanically, and it is the property to test before adopting this model.
Memorylessness, and Why It Is So Often Wrong
The exponential distribution satisfies P(X > s + t | X > s) = P(X > t) for every s and t. In words: given that you have already waited s units, the distribution of your remaining wait is identical to the distribution you started with. The clock resets continuously. It is the only continuous distribution with this property, and the memoryless panel in the tool demonstrates it numerically with your own numbers.
This is deeply counterintuitive and frequently false in practice. A bus that runs every ten minutes is not exponential; waiting nine minutes makes the next minute far more likely to produce a bus. A mechanical bearing is not exponential; a component that has run for five years is more likely to fail in the next hour than a new one. Human patience on a web page is not exponential either, since abandonment clusters at psychologically salient durations.
Where it does hold is where events have no mechanism connecting them to elapsed time. Radioactive decay is the canonical case, which is why our half-life calculator is really the same distribution expressed through its median. Arrivals at a busy service point over a short window are approximately memoryless, because the arrival of one customer carries no information about the next. The test is mechanical: ask whether anything about the system accumulates, ages, or builds up. If it does, the constant hazard assumption fails and the exponential will understate the risk of long waits, short waits, or both.
Fitting a Rate to Real Data
The maximum likelihood estimate of λ from a sample of waiting times is beautifully simple: it is the reciprocal of the sample mean. Add up all the waits, divide by the count, take the reciprocal, and that is your rate. No optimisation required.
Before you use it, run two checks. First, compare the sample standard deviation to the sample mean; they should be roughly equal, and a coefficient of variation far from 1 rules the exponential out immediately. Second, look at the fraction of observations below the mean, which should be close to 63.2 percent. Both checks take seconds and both catch the most common misfits. Bear in mind that the reciprocal-of-the-mean estimator is slightly biased upward in small samples, so a rate estimated from a dozen observations should not be quoted to three figures.
Note also that the exponential model has no free shape parameter. If the data are broadly right-skewed but the hazard clearly rises or falls with age, the usual generalisation is the Weibull distribution, which adds exactly that shape parameter and contains the exponential as the special case where the shape equals one.
Reading Percentiles Instead of Means
For a right-skewed distribution the mean is a poor summary of what to plan for, and the exponential is aggressively right-skewed. With a mean wait of 2 minutes, the 90th percentile is 4.61 minutes and the 99th is 9.21 minutes — more than four and a half times the mean. Capacity planned around the average will be wrong most of the times it matters.
The percentile output is there for that reason. The relationship xp = −ln(1−p) × mean gives a set of multipliers worth remembering: the 90th percentile is 2.30 times the mean, the 95th is 3.00 times, and the 99th is 4.61 times. Those ratios hold for every exponential distribution regardless of rate, which makes them a fast sanity check on any service level target. Our probability calculator handles the combinatorics if you then need the chance of several independent waits all staying under a threshold.
A Worked Example You Can Check Yourself
The defaults use a rate of 0.5 events per unit, so the mean wait is 2 units and the standard deviation is also 2. For a = 3, the survival probability is e−0.5×3 = e−1.5 = 0.22313, so the cumulative probability P(X ≤ 3) is 1 − 0.22313 = 0.77687, or 77.69 percent. The density at 3 is 0.5 × 0.22313 = 0.11157.
The interval probability between 3 and 5 is the difference of the two survival values: e−1.5 − e−2.5 = 0.22313 − 0.08209 = 0.14105, about 14.1 percent. The 90th percentile is −ln(0.10)/0.5 = 2.302585/0.5 = 4.6052, and the median is ln(2)/0.5 = 1.3863.
Now the memoryless check with 4 units already elapsed. The probability of waiting at least 3 more units given 4 have passed is e−0.5×7/e−0.5×4 = e−1.5 = 0.22313 — identical to the probability of waiting 3 units from scratch. Four units of waiting bought you nothing. Penn State's STAT 414 lesson on the exponential, gamma and chi-square distributions derives the same property, and shows how summing several exponential waits produces the gamma distribution rather than another exponential. If you need the continuous-growth arithmetic rather than the probability model, our exponential growth calculator handles that side.
Arb Digital's analysts size systems against the percentile that actually breaks, not the mean that looks comfortable.
Browse All Free Tools Talk To Our TeamCommon Mistakes to Avoid
- Entering a mean where a rate is expected — they are reciprocals, so the error is multiplicative and can be enormous. Use the toggle rather than converting by hand.
- Mixing time units — a rate per hour with values entered in minutes silently scales every answer by sixty.
- Assuming memorylessness for ageing systems — anything that wears, accumulates or runs to a schedule has a hazard rate that changes with time, and the exponential cannot represent it.
- Planning around the mean — more than 63 percent of exponential observations fall below the mean, and the tail runs far past it.
- Skipping the mean-equals-standard-deviation check — it takes one line and rules out the exponential immediately when the data does not fit.
Related Free Tools From Arb Digital
Count events in a fixed window with the Poisson distribution calculator, handle the discrete waiting case with the geometric distribution calculator, work with decay expressed through its median using the half-life calculator, compare against a symmetric model with the normal distribution calculator, combine independent events with the probability calculator, or project continuous growth with the exponential growth calculator. The free online tools hub lists every statistics tool we publish.
Frequently Asked Questions
It models the waiting time until the next event when events occur independently at a constant average rate. Common applications include time between arrivals in a queue, time to failure for components with a constant hazard rate, and radioactive decay.
They are reciprocals of each other. A rate of 0.5 events per hour corresponds to a mean waiting time of 2 hours. Entering one where the other is expected changes the answer substantially, so the tool lets you choose.
It means the remaining waiting time has the same distribution however long you have already waited. The chance of an event in the next minute never changes, and elapsed time carries no information about what happens next.
Both equal one divided by the rate, because the distribution has only one parameter controlling both its location and its spread. This makes a quick comparison of the sample mean and sample standard deviation a useful test of whether the model fits.
They describe the same process from two sides. If events arrive as a Poisson process with a given rate, the number of events in a fixed window is Poisson and the time between consecutive events is exponential with that same rate.
Take the reciprocal of the sample mean of the observed waiting times. That is the maximum likelihood estimate, though it carries a small upward bias when the sample is very small.
The Weibull distribution adds a shape parameter that allows the hazard rate to rise or fall with age, and it reduces to the exponential when that shape parameter equals one.
This page explains a statistical calculation for educational purposes only. It is not engineering, medical or financial advice, and the exponential model should not be applied to a real system without first checking that its constant-rate assumption holds.
