Calculator

Gamma Distribution

Calculate density, cumulative probabilities and quantiles of the gamma distribution with shape and scale.

Settings

Calculator focused on this distribution.

Result and visualization

Calculate PDF/PMF, CDF and percentiles instantly.

Result pending…

Explanation

The gamma distribution is a continuous distribution defined for positive values that generalizes the exponential distribution. While the exponential models the time until the first event of a Poisson process, the gamma with shape parameter k models the time until k consecutive events occur. This makes it ideal for cumulative waiting times, insurance claim severities and durations of processes with multiple stages. When k is a positive integer it is called the Erlang distribution. Its mean is k·θ and its variance is k·θ², so increasing θ shifts the distribution to the right and increasing k reduces the skewness, bringing it closer to a normal distribution via the central limit theorem.

Formula

\( f(x;k,\theta)=\frac{x^{k-1}e^{-x/\theta}}{\Gamma(k)\theta^k},\quad x>0 \)

Parameters

  • k: shape parameter (k > 0). Integer values correspond to the Erlang distribution.
  • θ (theta): scale parameter (θ > 0). Determines the spread and the mean together with k.
  • x: positive value at which the distribution is evaluated (x > 0).

Worked example

Situation: The repair time of an industrial machine is modeled with a Gamma distribution with shape parameter \(\alpha = 3\) and scale parameter \(\beta = 2\) hours. This is equivalent to assuming that the repair requires completing 3 independent stages, each with a mean duration of 2 hours.

Question 1: What is the mean repair time and what is the probability of finishing in 8 hours or less, \(P(X \leq 8)\)?

Solution: The mean of Gamma(\(\alpha, \beta\)) is: \[ E[X] = \alpha \cdot \beta = 3 \times 2 = 6 \text{ hours} \] To calculate \(P(X \leq 8)\) we use the regularized incomplete gamma function \(P(X \leq 8) = \Gamma_{\text{reg}}(3;\, 8/2) = \Gamma_{\text{reg}}(3;\, 4)\). Evaluating numerically: \[ P(X \leq 8) = 1 - e^{-4}(1 + 4 + 8) \approx 0.7619 \] There is approximately a 76.19% probability of completing the repair in 8 hours or less.

Question 2: What is the variance and what does it mean for the uncertainty of the repair time?

Solution: The variance of the Gamma is: \[ \text{Var}(X) = \alpha \cdot \beta^2 = 3 \times 4 = 12 \text{ h}^2 \implies \sigma = \sqrt{12} \approx 3.46 \text{ h} \] The standard deviation of 3.46 hours represents more than 57% of the mean, which reflects considerable variability in repair times.

Interpretation: With a mean of 6 hours and \(P(X \leq 8) \approx 0.762\), 23.8% of repairs take longer than 8 hours. If the service-level goal is to complete 90% of repairs on time, the 90th percentile of the Gamma(3, 2) can be used as the committed turnaround time: if the model is adequate, 90% of repairs would finish before that time and 10% would exceed it.

How to interpret the result

The calculator offers three types of output. The PDF, \( f(x) \), gives the probability density at point \( x \): it indicates how concentrated the distribution is in that area. With \( k = 1 \) the curve is decreasing (identical to the exponential); with \( k > 1 \) a peak appears (the mode is \( (k-1)\theta \)) and the curve takes on a skewed bell shape that shifts to the right as \( k \) increases. In the chart, the green area represents the cumulative probability of the selected range; the horizontal axis starts at zero (the gamma distribution only takes positive values).

The CDF, \( P(X \leq x) \), accumulates the probability from zero up to \( x \): it represents the probability that the variable (for example, a cumulative waiting time) is less than or equal to \( x \). If \( P(X \leq 1) = 0.78 \) with \( k = 3 \) and \( \theta = 0.25 \) hours, it means that 78% of processes complete their 3 stages in less than 1 hour. The right tail \( P(X > x) \) measures the probability that the process takes longer than \( x \) units, useful for sizing buffers and service times.

The percentile or quantile result returns the value \( x \) such that \( P(X \leq x) = p \). The 90th percentile indicates the time that only 10% of processes exceed, which allows setting service thresholds with a given coverage. In reliability engineering, the 10th percentile (B10) is the time before which 10% of components fail, a common standard for evaluating system durability.

Frequently asked questions

  • What do the shape and scale parameters mean? The shape controls the profile of the curve (low values give strong skew; high values make it close to normal) and the scale stretches the horizontal axis. The mean equals shape × scale.
  • How does it relate to the exponential and the chi-square? Both are special cases: the exponential is a gamma with shape 1, and the chi-square with k degrees of freedom is a gamma with shape k/2 and scale 2.
  • What is it used for in practice? For accumulated waiting times (time until the k-th event of a Poisson process), reliability, rainfall, insurance claim costs, and as a conjugate prior in Bayesian statistics.

Reference: Gamma distribution — Wikipedia