The gamma distribution has the pdf
\[f_X (x)=\frac{\beta^\alpha}{\color{orange}\Gamma(\alpha)} \color{red}{x^{\alpha - 1}\,e^{-\beta x}} \]
essentially the integrand of the gamma function:
\[\Gamma(z) = \int_0^\infty \color{red}{t^{z-1} \, e^{-t}}\;dt\]
This is explained here:
Let’s define a Poisson counting process with the number of arrivals at a certain time denoted as \(N(t)\).
As a Poisson process (i) \(N(0)=0\), the number of arrivals by time \(t\) (ii) follows a Poisson distribution,
\[N(t)=\text{Poi}(\lambda t),\]
which has a PMF
\[\frac{\left(\lambda t\right)^{n}\, e^{-\lambda t}}{n!}\]
where \(\lambda\) is the rate parameter in \(\text{Pois}(\lambda t)\) with \(n\in \mathbb N.\) Finally, (iii) additional condition is that the inter-arrival times \(X_1, X_2,\dots\) are iid.
The time to the first arrival is \(X_1\) while the time between the first and second arrival is \(X_2.\) Therefore, the inter-arrival time between arrival \(k-1\) and \(k\) is \(X_k.\)
Separately the time to the \(n\)-th arrival is \(S_n.\) So \(S_1\) is the time to the first arrival.
\[S_n = \sum_{k=1}^n X_k\]
such that \(S_1 = X_1,\) and \(S_2 = X_1 + X_2\) to \(S_n = X_1 + X_2 + \cdots + X_n.\)
The distribution of \(X_1\) is
\[\begin{align} F_{X_1}(x) &= \Pr(X_1 \leq x) \\ &= \Pr(\text{at least 1 arrival happens before time x})\\ &= \Pr(N(x) \geq 1)\\ &=1- \Pr(\text{no arrivals by time x})\\ &=1 - \Pr(N(x)=0)\\ &= 1 - \frac{\left(\lambda x\right)^0\, e^{-{\lambda x}}}{0!}\\ &= 1 - e^{-\lambda x} \end{align}\]
which is the CDF of the exponential distribution with rate \(\lambda\). So \(X_1 \sim \text{Exp}(\lambda)\).
Now the distribution of \(S_n\) (time to the \(n\)-th arrival) is
\[\begin{align} F_{S_n}(s)&=\Pr(S_n\leq s)\\ &=\Pr(n\text{ or more arrivals before time }s)\\ &=\Pr(N(s) \geq n)\\ &=\underset{\text{Pr(n or n+1 or n+2 ... infinite events before time s)}}{\underbrace{ \sum_{k=n}^{\infty} \frac{(\lambda s)^k\, e^{-\lambda s}}{k!}}} \end{align}\]
Notice that the CDF is an infinite sum of Poisson PMF’s.
The pdf of the time to the \(s\) arrival will be the derivative of this infinite sum of Poisson PMF’s.
\[\begin{align} f_{S_n}(s) &= \frac{d}{ds}F_{S_n}(s)\\ &=\sum_{k=n}^{\infty}\frac{d}{ds} \frac{(\lambda s)^k\, e^{-\lambda s}}{k!}\\ &= \sum_{k=n}^{\infty} \frac{k(\lambda s)^{k-1}\lambda\, e^{-\lambda s}}{k!} + \frac{(\lambda s)^{k}(-\lambda)\, e^{-\lambda s}}{k!}\\ &=\sum_{k=n}^{\infty} \frac{s^{k-1}\lambda^k\, e^{-\lambda s}}{(k-1)!} - \frac{s^{k}\lambda^{k+1}\, e^{-\lambda s}}{k!}\\ &= e^{-\lambda s}\left(\color{brown}{ \sum_{k=n}^{\infty}\frac{s^{k-1}\lambda^k}{(k-1)!}} -\color{magenta}{ \sum_{k=n}^{\infty} \frac{s^{k}\lambda^{k+1}}{k!}} \right)\\ \end{align} \]
Now the first term in the sum
\[\begin{align} \color{brown}{\sum_{k=n}^{\infty}\frac{s^{k-1}\lambda^k}{(k-1)!}}&=\frac{\lambda^n s^{n-1}}{(n-1)!} + \sum_{k=n+1}^{\infty}\frac{s^{k-1}\lambda^k}{(k-1)!}\\&=\frac{\lambda^n s^{n-1}}{(n-1)!} + \color{magenta}{ \sum_{k=n}^{\infty}\frac{s^{k}\lambda^{k+1}}{k!}} \end{align}\]
Therefore this last summand will be cancelled by the last term, resulting in
\[\begin{align} f_{S_n}(s)&=e^{-\lambda s} \frac{\lambda^n s^{n-1}}{(n-1)!}\\ &=\frac{\lambda^n}{(n-1)!}s^{n-1}\,e^{-\lambda s} \end{align}\]
which is the PDF of the gamma.
Notice that this expression can be remembered and thought of as just the structure of sort of the “derivative” of the Poisson PMF wrt to \(s\) (not exactly because the exponential is untouched):
\[\underset{\text{Poisson}}{\underbrace{\frac{(\lambda s)^k\;e^{-\lambda s}}{k!}}}=\frac{\lambda^k\; \color{red}{s^k}\;e^{-\lambda s}}{k!}\quad\quad\underset{\text{'d/ds'}}\to\quad \quad \lambda^k\quad \frac{\color{red}{k\, s^{k-1}}}{k!}\,e^{-\lambda s}= \lambda^k\quad \frac{\color{red}{ s^{k-1}}}{\color{magenta}{(k-1)!}}\,e^{-\lambda s}\]
For comparison and from Wikipedia the shape parameter is \(\alpha=n\) (the number of arrivals set to determine the probability that it will take different times \(s\) to reach the count \(n\)) and the rate parameter is \(\beta=\lambda\):
\[f_X (x)=\frac{\beta^\alpha}{\Gamma(\alpha)} x^{\alpha - 1}\,e^{-\beta x} \]
and \(\Gamma(\alpha)=(n-1)!\) with \(n\in \mathbb N.\)
The gamma distribution is explained here.
Factorials often appear in discrete probability distributions, such as the binomial and Poisson distributions. The gamma function’s connection to factorials means it can be used to derive continuous probability distributions, like the gamma and beta distributions. This ability to transition between discrete and continuous distributions enhances our toolkit in probability and statistics.
INTUITION: The pdf is just the integrand of the gamma function, and in it there is a polynomial form on the random variable, which tends to give the pdf an initial smooth upswing. Quickly, the exponential part of the pdf kicks in and kills the function. So they tend to have a positive skew. As in the case of the Weibull distribution with \(k=1\), the elimination of the polynomial part turns the distribution like an exponential. Here are some of the examples discussed below, showing this tendency:
[From ModelRisk David Vose, here]
The gamma distribution is connected to a number of different distributions:
\[f_X(x)=\frac{1}{2^{k/2}\color{orange}\Gamma(k/2)}\color{red}{x^{k/2-1}\,e^{-x/2}}\]
\[F_X(x) = \Pr(X<x)= \Pr(Y^2 < x)= \Pr(-\sqrt x \geq y \leq \sqrt x) \implies F_X(x)= F_Y(-\sqrt 2) - F_Y(\sqrt 2)\]
Since the pdf of the normal is \(\frac{1}{\sqrt{2\pi}}e^{-\frac{x^2}{2}}\), differentiating the expression above to get the equivalent with the pdf for \(\sqrt x\) results in:
\[f_X(x) = \frac 1 2 x^{-1/2} \frac{1}{\sqrt{2\pi}}e^{-\frac{x}{2}}+\frac 1 2 x^{-1/2} \frac{1}{\sqrt{2\pi}}e^{-\frac{x}{2}}= \frac{1}{\sqrt{2\pi}} x^{-1/2} e^{-\frac{x}{2}}\]
Since \(\Gamma(1/2)=\sqrt \pi\)
\[f_X(x) = \frac{\frac{1}{2}^{1/2}}{\Gamma(1/2)}x^{-1/2} e^{-\frac{x}{2}} \sim \Gamma(1/2,1/2)\]
which is a gamma distribution with \(\lambda = 1/2\) and \(\alpha = 1/2\) comparing to the form of the gamma:
\[f_X(x) = \frac{\lambda^\alpha}{\Gamma(\alpha)}x^{\alpha -1}e^{-\lambda x}\] Naturally, it is also distributed as a \(\chi^2(1 \text{df})\) (see above).
\[f_X(x)= \frac{\color{orange}\Gamma\left(\frac{\nu + 1}{2}\right)}{\sqrt{\pi \nu}\,\color{orange}\Gamma\left(\frac{\nu}{2}\right)}\left( 1 + \frac{x^2}{\nu}\right)^{-\frac{\nu+1}{2}}\]
\[f_X (x)=\frac{\beta^\alpha}{\Gamma(\alpha)}\color{red}{x^{\alpha - 1}\,e^{-\beta x}} =\beta\,e^{-\beta x}\]
\[f_X(x)=\frac{\lambda^k\,\color{red}{x^{k-1}\,e^{-\lambda x}}}{(k-1)!}\]
with \(\lambda\) being the rate parameter, \(k\in\{1,2,\dots\}\) the shape parmeter
\[f_X(x)=\frac{k}{\lambda}\color{red}{\left(\frac x k\right)^{k-1}\, e^{-(x/\lambda)^k}}\]
\[f_X(x)=\frac{x^{\alpha -1}(1-x)^{\beta-1}}{\frac{\color{orange}\Gamma(\alpha)\color{orange}\Gamma(\beta)}{\color{orange}\Gamma(\alpha +\beta)}}\]
\[f_X(x) = \sqrt{\frac{c}{2\pi}}\frac{e^{\frac{-c}{2(x-\mu)}}}{(x-\mu)^{3/2}}=\sqrt{\frac{c}{2\pi}}\color{red}{(x-\mu)^{-3/2}\,e^{\frac{-c}{2(x-\mu)}}}\]
If \(X\) is a random variable following a gamma distribution with shape parameter α and rate parameter β, then the random variable \(Y = 1/X\) follows a Lévy distribution with parameter α.
\[f_X(x) =\frac{1}{\sigma}\color{red}{t(x)^{\xi+1}\,e^{-t(x)}}\]
where if \(\xi \neq 0\)
\[t(x) = \left(1 + \xi\left(\frac{x-\mu}{\sigma} \right)\right)^{-1/\xi}\]
and if \(\xi = 0\)
\[t(x) = \exp\left(-\frac{x -\mu}{\sigma}\right)\]
\[f_X(x) = \color{red}{\frac{x}{\sigma^2}\,e^{-x^2/2\sigma^2}}\]
\[X = \frac{U_1/u}{U_2/v}\] The pdf is (reference):
\[f_F(f) = \frac{\color{orange}\Gamma\left( \frac{u+v}{2} \right)}{\color{orange}\Gamma\left( \frac{u}{2} \right) \cdot \color{orange}\Gamma\left( \frac{v}{2} \right)} \cdot \left( \frac{u}{v} \right)^{\frac{u}{2}} \cdot f^{\frac{u}{2}-1} \cdot \left( \frac{u}{v}f+1 \right)^{-\frac{u+v}{2}} \]
The gamma distribution is used in:
Survival Analysis: In survival analysis, the gamma distribution can be used to model the time to an event of interest, such as the time to failure of a component or the time to death.
Insurance: The gamma distribution is used in insurance modeling to represent the distribution of claim amounts.
Engineering: In engineering, the gamma distribution is used to model various processes, such as the time between failures of equipment or the distribution of material properties.
Waiting Times: The gamma distribution is often used to model waiting times between events, especially when the events are not independent or when the waiting time is not exponentially distributed.
The “Penelope distributions”:
Exponential: Models the inter-arrival time in a Poisson process. Waiting times constant & memoryless. The continuous time analogue of the geometric distribution.
Erlang: It is a discretized form of the gamma (check the denominator, and the fact that the shape parameter \(k\) has to be a positive integer)
\[f_X(x)=\frac{\lambda^k\,\color{red}{x^{k-1}\,e^{-\lambda x}}}{(k-1)!}\] It is used to model the sum of independent exponential random variables: distribution of the waiting time until the \(k\)-th event of a Poisson process of rate \(\lambda\).
If \(k = 1\) it becomes the exponential distribution.
\[f_X(x)= \frac{k}{\lambda}\left(\frac{x}{\lambda}\right)^{k-1}\,e^{-(x/\lambda)^k}\]
So the exponential \(e^{-\left(\frac{x}{\lambda}\right)^k}\) has a polynomial shape!
Therefore the Weibull will be used to model time-to-failure:
A value of \(k < 1\) indicates that the failure rate decreases over time (like in case of the Lindy effect, which however corresponds to Pareto distributions[7] rather than Weibull distributions).
A value of \(k = 1\) indicates that the failure rate is constant over time. This might suggest random external events are causing mortality, or failure. The Weibull distribution reduces to an exponential distribution.
A value of \(k > 1\) indicates that the failure rate increases with time. This happens if there is an “aging” process, or parts that are more likely to fail as time.
\[T_n = \sum_{i=1}^n X_i \sim \text{Gamma}(n, \lambda)\]
were \(X_i\) are iid exponentials.
The gamma is the continuous-time analogue of the negative binomial.
NOTE: These are tentative notes on different topics for personal use - expect mistakes and misunderstandings.