Download as DOC, PDF, TXT or read online from Scribd
Download as doc, pdf, or txt
You are on page 1/ 30
Gamma distribution
From Wikipedia, the free encyclopedia
Not to be confused with Gamma function.
Gamma
Probability density function
Cumulative distribution function
k > 0 shape > 0 shape
Parameters
> 0 scale > 0 rate
Support
Probability
density
function (pdf)
Cumulative
distribution
function (CDF)
Mean
(see digamma function) (see digamma function)
Median No simple closed form No simple closed form
Mode
Variance
(see trigamma function) (see trigamma function)
Skewness
Excesskurtosis
Entropy
Moment-
generating
function (mgf)
Characteristic
function
In probability theory and statistics, the gamma distribution is a two-parameter family of continuous probability distributions. There are three different parametrizations in common use:
1. With a shape parameter k and a scale parameter .
2. With a shape parameter = k and an inverse scale parameter = 1/,
called a rate parameter.
3. With a shape parameter k and a mean parameter = k/.
In each of these three forms, both parameters are positive real numbers.
The parameterization with k and appears to be more common in econometrics and certain other applied fields, where e.g. the gamma distribution is frequently used to model waiting times. For instance, in life testing, the waiting time until death is a random variable that is frequently modeled with a gamma distribution. [1]
The parameterization with and is more common in Bayesian statistics, where the gamma distribution is used as a conjugate prior distribution for various types of inverse scale (aka rate) parameters, such as the of an exponential distribution or a Poisson distribution or for that matter, the of the gamma distribution itself. (The closely related inverse gamma distribution is used as a conjugate prior for scale parameters, such as the variance of a normal distribution.)
If k is an integer, then the distribution represents an Erlang distribution; i.e., the sum of k independent exponentially distributedrandom variables, each of which has a mean of (which is equivalent to a rate parameter of 1/).
The gamma distribution is the maximum entropy probability distribution for a random variable X for which E[X] = k = / is fixed and greater than zero, and E[ln(X)] = (k) + ln() = () ln() is fixed ( is the digamma function).[2]
Contents
[hide]
1 Characterization using shape k and scale
o 1.1 Probability density function
o 1.2 Cumulative distribution function
2 Characterization using shape and rate
o 2.1 Probability density function
o 2.2 Cumulative distribution function
3 Properties
o 3.1 Skewness
o 3.2 Median calculation
o 3.3 Summation
o 3.4 Scaling o 3.5 Exponential family
o 3.6 Logarithmic expectation
o 3.7 Information entropy
o 3.8 KullbackLeibler divergence
o 3.9 Laplace transform
4 Parameter estimation
o 4.1 Maximum likelihood estimation
o 4.2 Bayesian minimum mean-squared error
5 Generating gamma-distributed random variables
6 Related distributions
o 6.1 Special cases
o 6.2 Conjugate prior
o 6.3 Compound gamma
o 6.4 Others
7 Applications
8 Notes
9 References
10 External links
Characterization using shape k and scale [edit]
A random variable X that is gamma-distributed with shape k and scale is denoted
Probability density function[edit]
Illustration of the gamma PDF for parameter values over k andx with set to 1, 2, 3, 4, 5 and 6. One can see each layer by itself here [1] as well as by k [2] and x. [3].
The probability density function using the shape-scale parametrization is
Here (k) is the gamma function evaluated at k.
Cumulative distribution function[edit]
The cumulative distribution function is the regularized gamma function:
where (k, x/) is the lower incomplete gamma function.
It can also be expressed as follows, if k is a positive integer (i.e.,
the distribution is an Erlang distribution):[3]
Characterization using shape and rate
[edit]
Alternatively, the gamma distribution can be parameterized in
terms of a shape parameter = k and an inverse scale parameter = 1/, called a rate parameter. A random variable X that is gamma-distributed with shape and rate is denoted
Probability density function[edit]
The corresponding density function in the shape-rate parametrization is
Both parametrizations are common because either
can be more convenient depending on the situation.
Cumulative distribution function[edit] The cumulative distribution function is the regularized gamma function:
where (, x) is the lower incomplete gamma
function.
If is a positive integer (i.e., the distribution is
an Erlang distribution), the cumulative distribution function has the following series expansion:[3]
Properties[edit] Skewness[edit] The skewness is equal to , it depends only on the shape parameter (k) and approaches a normal distribution when k is large (approximately when k > 10). Median calculation[edit] Unlike the mode and the mean which have readily calculable formulas based on the parameters, the median does not have an easy closed form equation. The median for this distribution is defined as the value such that
A formula for approximating the
median for any gamma distribution, when the mean is known, has been derived based on the fact that the ratio /( ) is approximately a linear function of kwhen k 1.[4] The approximation formula is
where is the mean.
Summation[edit] If Xi has a Gamma(ki, ) distribution for i = 1, 2, ..., N (i.e., all distributions have the same scale parameter ), then
provided all Xi are independent.
For the cases where
the Xi are independent but have different scale parameters see Mathai (1982) and Moschopoulos (1984).
The gamma distribution
exhibits infinite divisibility.
Scaling[edit] If
then for any c > 0,
Hence the use of
the term "scale parameter" to describe .
Equivalently, if
then for any c > 0,
Hence the use of the term "inverse scale parameter" to describe .
Expone ntial family[e dit] The gamma distribution is a two- parameter exponentia l family with natural parameter s k 1 and 1/ (equivalent ly, 1 and ), and natural statistics X and ln(X).
If the shape parameter k is held fixed, the resulting one- parameter family of distribution s is a natural exponentia l family.
Logarit hmic expect ation[ed it] One can show that
or equiva lently,
w h er e
is th e di g a m m a fu n cti o n.
T hi s c a n b e d er iv e d u si n g th e e x p o n e nt ial fa m ily fo r m ul a fo r th e m o m e nt g e n er at in g fu n cti o n of th e s uf fic ie nt st at ist ic, b e c a u s e o n e of th e s uf fic ie nt st at ist ic s of th e g a m m a di st ri b ut io n is ln (x ).
I n f o r m a ti o n e n tr o p y [ e di t] T h e in fo r m at io n e nt ro p yi s In th e k,
pa ra m et eri za tio n, th ei nf or m ati on en tro py is gi ve n by
Kull back
Leibl er diver genc e[edit ]
Illustration of the Kullback Leibler (KL) divergence for two gamma PDFs. Here = 0 + 1 which are set to 1, 2, 3, 4, 5 and 6. The typical asymmetry for the KL divergence is clearly visible.
The Ku llback Leibler diverg ence ( KL- diverg ence), as with the inform ation entrop y and various other theoret ical propert ies, are more comm only seen using the ,
param eteriza tion becaus e of their uses in Bayesi an and other theoret ical statisti cs frame works.
The KL- diverg ence of Gamm a(p, p) ("true" distribu tion) from Gamm a(q, q) ("appro ximatin g" distribu tion) is given by[5]
Written using the k,
parameteriz ation, the KL- divergence of Gamma(kp, p) from Gamma(kq, q) is given by
Laplace transform[ edit] The Laplace transform of the gamma PDF is
Parameter estimation[e t]
Maximum likelihood estimation[e ] The likelihood function for N iid observati s (x1, ..., xN) is
from which we ca the log-likelihood Finding the maxim respect to by tak derivative and set to zero yields the likelihood estimato parameter:
Substituting this in likelihood function
Finding the maxim
to k by taking the it equal to zero yie
There is no closed function is numeri so if a numerical s be found using, fo method. An initial either using the m using the approxim
If we let
then k is approxim which is within 1.5 form for the Newto is:[7]
Bayesian mi error[edit] With known k and for theta (using th
Denoting
Integration over revealing that 1/
The moments can
which shows that
distribution for the
Generating
Given the scaling
1 as we can later
Using the fact tha
and noting the me if U is uniformly di using the "-addit
where Uk are all u
generate a variab property once mo
Random generatio none are uniforml parameter, the alg parameter, one ca Algorithm GD (sha and Feast Algorith
The following is a
1. Let m be 1.
2. Generate V3m
3. If
4. Let
5. Let
6. If
7. Assume =
A summary of this
where
is the inte
has been ge Uk and Vl are
While the above a
of k and in genera or table-based me
Related dist Special case
If X ~ Gamma
If X ~ Gamma distribution w constant, then
If k is an integ distribution of
process with then
If X has a Ma .
X ~ Gamma(k parameters p
an exponentia
Conjugate p In Bayesian infere distributions: the P known shape ,in parameter.
The gamma distri
where Z is the nor can be found by u
where n is the num
Compound g If the shape param unknown, then a g distribution, which the compound ga
Others[edit]
If X ~ Gamma parameter k a distribution.
If X ~ Gamma distribution w
If Xi are indep where S = X1
For large k th variance 2 =
The gamma d known mean.
The Wishart d positive-defin The gamma d integer gamm
Among the di discrete analo
Tweedie distr dispersion mo
Applications
The gamma distri
means that aggre modelled by a gam level Poisson regr distribution is a ne
In neuroscience, t intervals.[19] Althou biophysical motiva
In bacterial gene e gamma distributio bursts per cell cyc lifetime.[20]
The gamma distri
prior for the precis for the exponentia Notes[edit]
1. Jump up^ Se
2. Jump up^ Pa
heteroskedast
3. ^ Jump up to:a
4. Jump up^ Ba
distribution". V
5. Jump up^ W.
Gamma, Dirich
6. Jump up^ Min
us/um/people/
7. Jump up^ Ch
Distribution an
8. Jump up^ Lu
Chapter 9, Se
9. Jump up^ De
10. ^ Jump up to:a
technique. Co
11. Jump up^ Ah
binomial distrib
12. Jump up^ Ch
290-295.
13. Jump up^ Ma
321-325.
14. Jump up^ Lu
Chapter 9, Se
15. Jump up^ Fin
enhancement 16. Jump up^ Du
31. doi:10.100
17. Jump up^ p.
CRC 2007
18. Jump up^ Ak
24, 419 428
19. Jump up^ J. G
cells of the cat
20. Jump up^ N.
An analytical f
References
R. V. Hogg an Macmillan. (S
P. G. Moscho variables, An
A. M. Mathai Statistical M
External link
Hazewinkel, M Mathematics,
Weisstein, Er
Engineering S Categories: Continuous d Factorial and Conjugate pri Exponential fa Infinitely divis
Navigatio Create accoun
Log in Article Talk Read Edit View history
Main page Contents Featured cont Current events Random articl Donate to Wik
Interaction Help About Wikiped Community po Recent chang Contact page Tools Print/export Languages
