Professional Documents
Culture Documents
c
2008 American Statistical Association DOI: 10.1198/000313008X270448 The American Statistician, February 2008, Vol. 62, No. 1 45
Devroye (2006) refers to Dickman’s, Kolmogorov–Smirnov, • The minimum property (M) indicates that the smallest of
Kummer’s, Linnik–Laha, theta, and de la Vallée–Poussin dis- independent and identically distributed random variables
tributions in his chapter on variate generation. from a distribution comes from the same distribution fam-
ily.
2. DISTRIBUTION PROPERTIES Example: If X i ∼ exponential(αi ) for i = 1, 2, . . . , n, and
X 1 , X 2 , . . . , X n are independent, then
There are several properties that apply to individual distribu-
tions listed in Figure 1. n
X !
min{X 1 , X 2 , . . . , X n } ∼ exponential 1 (1/αi ) .
• The linear combination property (L) indicates that lin- i=1
ear combinations of independent random variables having
this particular distribution come from the same distribution • The maximum property (X) indicates that the largest of
family. independent and identically distributed random variables
Example: If X i ∼ N μi , σi2 for i = 1, 2, . . . , n;
from a distribution comes from the same distribution fam-
a1 , a2 , . . . , an are real constants, and X 1 , X 2 , . . . , X n are ily.
independent, then Example: If X i ∼ standard power (βi ) for i = 1, 2, . . . , n,
and X 1 , X 2 , . . . , X n are independent, then
n n n
!
ai X i ∼ N ai μi , ai2 σi2 n
X X X !
max{X 1 , X 2 , . . . , X n } ∼ standard power
. X
i=1 i=1 i=1 βi .
i=1
• The convolution property (C) indicates that sums of inde-
pendent random variables having this particular distribu- • The forgetfulness property (F), more commonly known as
tion come from the same distribution family. the memoryless property, indicates that the conditional dis-
Example: If X i ∼ χ 2 (n i ) for i = 1, 2, . . . , n, and X 1 , tribution of a random variable is identical to the uncondi-
X 2 , . . . , X n are independent, then tional distribution. The geometric and exponential distri-
butions are the only two distributions with this property.
n n
This property is a special case of the residual property.
!
Xi ∼ χ 2
ni .
X X
i=1 i=1 • The residual property (R) indicates that the conditional
distribution of a random variable left-truncated at a value
• The scaling property (S) implies that any positive real in its support belongs to the same distribution family as the
constant times a random variable having this distribution unconditional distribution.
comes from the same distribution family.
Example: If X ∼ Uniform(a, b), and k is a real constant
Example: If X ∼ Weibull(α, β) and k is a positive, real satisfying a < k < b, then the conditional distribution of
constant, then X given X > k belongs to the uniform family.
k X ∼ Weibull(αk β , β).
• The variate generation property (V) indicates that the in-
verse cumulative distribution function of a continuous ran-
• The product property (P) indicates that products of inde- dom variable can be expressed in closed form. For a dis-
pendent random variables having this particular distribu- crete random variable, this property indicates that a variate
tion come from the same distribution family. can be generated in an O(1) algorithm that does not cycle
Example: If X i ∼ lognormal(μi , σi2 ) for i = 1, 2, . . . , n, through the support values or rely on a special property.
and X 1 , X 2 , . . . , X n are independent, then Example: If X ∼ exponential(α), then
46 Teacher’s Corner
Figure 1. Univariate distribution relationships.
tributions. Consider the relationship between the normal distri- where X 1 (δ), X 2 (γ ) are noncentral chi-square random
bution and the standard normal distribution. If X ∼ N (μ, σ 2 ), variables with n 1 , n 2 degrees of freedom, respectively,
then X −μ ∼ N (0, 1) as indicated on the chart. Conversely, if (Johnson, Kotz, and Balakrishnan 1995, p. 480).
X ∼ N (0, 1), then μ + σ X ∼ N (μ, σ 2 ). The first direction
σ
of the transformation is useful for standardizing random vari- • A normal and uniform random variable are special and lim-
ables to be used for table lookup, while the second direction iting cases of an error random variable (Evans, Hastings,
is useful for variate generation. In most cases, though, an in- and Peacock 2000, p. 76).
verse transformation is implicit and is not listed on the chart for • A binomial random variable is a special case of a power se-
brevity (e.g., extreme value random variable as the logarithm of ries random variable (Evans, Hastings, and Peacock 2000,
a Weibull random variable and Weibull random variable as the p. 166).
exponential of an extreme value random variable).
Several of these relationships hint at further distributions that • The limit of a von Mises random variable is a normal ran-
have not yet been developed. First, the extreme value and log dom variable as κ → ∞ (Evans, Hastings, and Peacock
gamma distributions indicate that the logarithm of any survival 2000, p. 191).
48 Teacher’s Corner
• The half-normal, Rayleigh, and Maxwell–Boltzmann dis- • Muth’s distribution: Muth (1977)
tributions are special cases of the chi distribution with
n = 1, 2, and 3 degrees of freedom (Johnson, Balakrish- • negative hypergeometric distribution: Balakrishnan and
nan, and Kotz 1994, p. 417). Nevzorov (2003), Miller and Fridell (2007)
• A function of the ratio of two independent generalized • power distribution: Balakrishnan and Nevzorov (2003)
gamma random variables has the beta distribution (Stacy
• TSP distribution: Kotz and van Dorp (2004)
1962).
• Zipf distribution: Ross (2006).
Additionally, there are transformations where two distribu-
tions are combined to obtain a third, which were also omitted to
maintain a planar graph. Two such examples are: A. APPENDIX: PARAMETERIZATIONS
n2 1
f (x) = x n/2−1 e−x/2 , x >0
,
2n/2 0(n/2)
x = max(0, n 1 + n 2 − n 3 ), . . . , n 2
Doubly noncentral F:
Pascal (negative binomial):
j k
n−1+x n e e
−δ/2 1 δ −γ /2 1 γ
f (x) = p (1 − p)x , x = 0, 1, . . . f (x) =
2 2
∞
XX ∞
x j! k!
j=0 k=0
Poisson (μ > 0): (n 1 /2)+ j (n 2 /2)+k (n 1 /2)+ j−1
×n 1 n2 x
μx e−μ
f (x) = x = 0, 1, . . .
x! ×(n 2 + n 1 x)− 2 (n 1 +n 2 )− j−k
, 1
Polya: 1 1
−1
× B n 1 + j, n 2 + k x >0
2 2
,
x−1 n−x−1 n−1
n Y
f (x) = ( p + jβ) (1 − p + kβ) (1 + iβ),
Y Y
x Doubly noncentral t:
j=0 k=0 i=0
See Johnson, Kotz, and Balakrishnan (1995, p. 533)
x = 0, 1, . . . , n
Erlang:
Power series (c > 0; A(c) = ax c x ):
P
x
1
ax c x f (x) = x n−1 e−x/α , x >0
f (x) = x = 0, 1, . . . α n (n − 1)!
A(c)
,
Rectangular (discrete uniform, n = 0, 1, . . .): Error (exponential power, general error; −∞ < a < ∞, b >
0, c > 0):
1
f (x) = x = 0, 1, . . . , n exp −(|x − a|/b)2/c /2
n+1 f (x) = −∞ < x < ∞
,
b(2c/2+1 )0(1 + c/2)
,
Zeta:
1
f (x) = x = 1, 2, . . . Exponential (negative exponential):
xα
P∞ ,
i=1 (1/i)
f (x) = (1/α)e−x/α , x >0
α
Zipf (α ≥ 0):
1 Exponential power:
f (x) = x = 1, 2, . . . , n
xα
Pn ,
i=1
f (x) = (e1−e )eλx λκ x κ−1 , x >0
(1/i)α λx κ κ
50 Teacher’s Corner
Gamma–normal: Log logistic:
See Evans, Hastings, and Peacock (2000, p. 103)
f (x) = x >0
λκ(λx)κ−1
Generalized gamma: [1 + (λx)κ ]2
,
1 1
f (x) = √ exp − (log(x/α)/β)2 , x >0
α γβ 0(β)
Generalized Pareto: 2π βx 2
Logistic:
f (x) = γ + (1 + x/δ)−κ e−γ x , x >0
κ
x +δ λκ κeκ x
f (x) = −∞ < x < ∞
Gompertz (κ > 1): [1 + (λe x )κ ]2
,
Logistic–exponential:
f (x) = δκ x e[−δ(κ x >0
x −1)/ log κ]
,
αβ(eαx − 1)β−1 eαx
Hyperbolic–secant: f (x) = x >0
(1 + (eαx − 1)β )2
,
x β−1 (1 + x)−β−γ
2 −1
f (x) = 2 2
x >0
∞
X
f (x) = x >0 k! 2
n+2k
0( n+2k
∙
B(β, γ )
,
k=0 2
2 )
,
1 22i (i!)2
i=0
f (x) = −∞ < x < ∞
π(1 + x 2 )
,
[Received October 2007. Revised December 2007.]
Standard normal:
e−x /2 REFERENCES
2
52 Teacher’s Corner
(1995), Continuous Univariate Distributions (Vol. II, 2nd ed.), New Patel, J.K., Kapadia, C.H., and Owen, D.B. (1976), Handbook of Statistical
York: Wiley. Distributions, New York: Marcel Dekker, Inc.
Jones, M.C. (2002), “The Complementary Beta Distribution,” Journal of Statis- Patil, G.P., Boswell, M.T., Joshi, S.W., and Ratnaparkhi, M.V. (1985), Discrete
tical Planning and Inference, 104, 329–337. Models, Burtonsville, MD: International Co-operative Publishing House.
Jordan, C.W. (1967), Life Contingencies, Chicago: Society of Actuaries. Patil, G.P., Boswell, M.T., and Ratnaparkhi, M.V. (1985), Univariate Con-
Kotz, S., and van Dorp, J.R. (2004), Beyond Beta: Other Continuous Families tinuous Models, Burtonsville, MD: International Co-operative Publishing
of Distributions with Bounded Support and Applications, Hackensack, NJ: House.
World Scientific. Prentice, R.L. (1975), “Discrimination Among Some Parametric Models,”
Lan, L., and Leemis, L. (2007), “The Logistic–Exponential Survival Distribu- Biometrika, 62, 607–619.
tion,” Technical Report, The College of William & Mary, Department of Ramberg, J.S., and Schmeiser, B.W. (1974), “An Approximate Method for Gen-
Mathematics. erating Asymmetric Random Variables,” Communications of the Associa-
Leemis, L. (1986), “Relationships Among Common Univariate Distributions,” tion for Computing Machinery, 17, 78–82.
The American Statistician, 40, 143–146. Ross, S. (2006), A First Course In Probability (7th ed.), Upper Saddle River,
Marshall, A.W., and Olkin, I. (1985), “A Family of Bivariate Distributions Gen- NJ: Prentice Hall.
erated by the Bivariate Bernoulli Distribution,” Journal of the American Sta- (2007), Introduction to Probability Models (9th ed.), New York: Aca-
tistical Association, 80, 332–338. demic Press.
McDonald, J.B. (1984), “Some Generalized Functions for the Size Distribution Seshadri, V. (1993), The Inverse Gaussian Distribution, Oxford: Oxford Uni-
of Income,” Econometrica, 52, 647–663. versity Press.
Miller, G.K., and Fridell, S.L. (2007), “A Forgotten Discrete Distribution? Re-
Shapiro, S.S., and Gross, A.J. (1981), Statistical Modeling Techniques, New
viving the Negative Hypergeometric Model,” The American Statistician, 61,
York: Marcel Dekker.
347–350.
Smith, R.M., and Bain, L.J. (1975), “An Exponential Power Life-Testing Dis-
Muth, E.J. (1977), “Reliability Models with Positive Memory Derived from the
tribution,” Communications in Statistics, 4, 469–481.
Mean Residual Life Function” in The Theory and Applications of Relia-
bility, eds. C.P. Tsokos and I. Shimi, New York: Academic Press, Inc., pp. Song, W.T. (2005), “Relationships Among Some Univariate Distributions,” IIE
401–435. Transactions, 37, 651–656.
Nakagawa, T., and Yoda, H. (1977), “Relationships Among Distributions,” Stacy, E.W. (1962), “A Generalization of the Gamma Distribution,” Annals of
IEEE Transactions on Reliability, 26, 352–353. Mathematical Statistics, 33, 1187–1192.
Ord, J.K. (1972), Families of Frequency Distributions, New York: Hafner Pub- Taha, H.A. (1982), Operations Research: An Introduction (3rd ed.), New York:
lishing. Macmillan.