EN202
SU13
Homework 1
08/07/2013
Varun Munjal
504109452
Gamma Distribution
Introduction
Gamma distribution is a two parameter probability model used in situations where the time gap between events is irrelevant. Events that follow Poisson statistics are a good example of such distributions. E.g. to model the amount of rain drops collected at a certain time given that the arrival of rain drops follow Poisson statistics.
Probability Density Function
For a random variable X, and parameters k and , the probability density function of a Gamma distribution is given by: , Where,
1 / k
k = number of occurrences of an event, also known as the shape factor (k) = (k-1)! Is the Gamma function = = mean number of events per time unit = 1/, where is the mean time between events. Also known as the scaling factor
Normalization
For positive values of k and , the probability density function of the Gamma distribution is properly normalized Proof: Find the integral from 0 to of the probability density function.
Replace gamma function with its actual definition:
Substitute / in the numerator
Which is simply:
=1 This proves that the probability density function is properly normalized.
Special cases
Relationship with Chi-square distribution Chi square distribution for a random variable Y for n degrees of freedom is given by:
n 2
In the gamma distribution function, Substituting k = n/2 and = 2
1 , 2 / n 2 2 2
So it can be seen that Gamma distribution is the same as Chi sq distribution for k = n/2 and = 2 Relationship with Exponential Distribution: A generic exponential distribution function for random variable Y with parameter is given by: 1 /
In the gamma distribution function, Substituting k = 1 and = 1, 1 1 /
1,
So it can be seen that Gamma distribution is the same as exponential distribution for k = 1 and =
1 /
Cumulative Distribution
The cumulative distribution function of the gamma distribution is the integral: , , Substitute x = t. The above integral reduces to: , ,
1 k
, ,
Where is called the incomplete gamma function.
There is no closed form solution for the cumulative distribution and can be evaluated numerically by using GAMMADIST function in MS Excel. Usage: , , , false , , , true
Examples:
Probability Density Function
0.3
=3
0.25 0.2 f(x) 0.15 0.1 0.05 0 0 5 10 15 X 20 25 30 35 k=1 k=2 k=3
One can see that for k=1, the distribution is exponential
Cumulative Distribution Function
1.2
0.8
f(x)
0.6
k=1 k=2
0.4
k=3
0.2
0 0 5 10 15 X 20 25 30 35
Reliability function
The reliability function is evaluated using MS Excel. 1 , , , true
Failure function
Failure function is evaluated using MS Excel. , , , true
Failure rate
, , , false 1 , , , true
Inverse function
The probability density function can be found by computing the inverse of the failure function. , ,
Moment generating function
The moment generating function can be derived by evaluating the integral: , ,
Substitute , ,
, ,
, ,
1 / / k k1 y
Then,
, ,
The integral is same as the gamma function. Therefore, , ,
1 1 y
1 1 y
Higher order moments
First moment: Differentiate the moment generating function with respect to y once and set y = 0.
Mean:
Median:
Since the cumulative distribution function has no closed form, the median expression does not have a closed form either. It can be evaluated using MS Excel Mode: Differentiate the probability density function f(x) with respect to x and find the maxima. 0 k 1 0 k 1
For k =1, the distribution is exponential.
For k >1,
Second moment:
Differentiate the moment generating function with respect to y twice and set y = 0. 1
Variance:
Third moment: Differentiate the moment generating function with respect to y three times and set y = 0. 1 2
Skewness:
1 3 2 2
Since k is always positive, the gamma distribution function is always positively skewed
Independent Gamma distributions
Sum of n independent Gamma distributions can be shown to be Gamma distributed if they have the same scaling parameter . Proof: The moment generating function for n independent Gamma distributions with shape parameters k1, k2, k3 kn is the product of the individual moment generating functions. Let M(y,k) be the product. , 1 1 1 . 1 1 1 1 1
which is of the same form as the moment generating function of a Gamma distribution. The shape factor of the sum of n independent gamma distributions with the same scaling factor is the sum of the individual shape factors.
References
1. Richard J Larsen, Morris L Marx (1986). Mathematical Statistics and its Applications, PrenticeHall, Chapter 4, pages 226-229 2. C. Radhakrishna Rao (1952). Advanced Statistical Methods in Biometric Research, Wiley, Chapter 2, pages 40-41 3. Robert V. Hogg, Joseph W. McKean, Allen Thornton Craig (2005). Introduction to Mathematical Statistics, Pearson Education, Chapter 2 4. Christian Walck (2007). Hand-book on Statistical Distributions for Experimentalists, University of Stockholm, Chapter 17, pages 69-71 5. Wikipedia, http://en.wikipedia.org/wiki/Gamma_distribution
