You are on page 1of 2

Joint cdf F a , b=P {X a ,Y b } X,Y independent, pdf fx, fy pdf of X+Y

marginal cdf F X a=P { X a }=F a ,

f X Y a= f X a y f Y y dy
discrete pmf p x , y ={ X = x , Y = y }

marginal pmf p X x = p x , y X,Y independent gamma with (s,), (t, ) X+Y gamma with (s+t, )
y: p x , y0 X,Y independent normal with (x,2x), (y,2y)
X+Y normal with (x+y, 2x +2y)
continuous pdf P { X ,Y C }= f x , y dA X,Y independent Poisson with 1, 2 X+Y Poisson with 1+2
C
2
X,Y independent binomial with (n,p),(m,p) X+Y binomial with (n+m,p)
f a ,b= F a , b X1,....,Xn independent exponential all with
xy X1+...+Xn gamma with (n,)
Z1,...,Zn independent unit normal Y=Z12 + + Zn2 2 with n deg. free.
marginal pdf f X x = f x , y dy Z1,...,Zn independent normal with (,)
2

n
X i
multinomial distribution
n! n n n
p p ... p r 1 2 r
Y=
2 with n deg. free.
i=1
n1!n2!...n r! 1 2 discrete conditional distribution
independence P { X A , Y B }=P { X A}P {Y B} conditional pmf p X Y xy= p x , y/ p Y y
cdf, pmf, pdf , joint pdf/pmf h(x),g(y) conditional cdf
gamma distribution with parameter (t,)
F XY x , y= P { X xY = y }= p X Y ay
pdf f x = e x x t1 / t a x
a continuous conditional distribution
cdf F x = e x xt1 / t conditional pdf f X Y xy = f x , y/ f Y y
0 a
mean E [ X ]=t / conditional cdf F XY a , y= f xy dx
X Y
variance Var X =t / 2
joint pdf of function of random variable :
gamma function t= e y
y t1
dy (1) given : r.v. X1,X2 with joint pdf fX1X2 X Y ()
0
Y1=g1(X1,X2), Y2=g2(X1,X2)
t=t1 t1 , n=n1! 1
(2) fY Y2 y1, y 2 = f X X2 x1, x 2
1/2=
1 1
J x 1, x 2
incomplete gamma function with parameters (x,t) mean of function of two jointly distributed random variables
x (1) Given : r.v. X,Y, with joint pmf/pdf f(x,y) then
x ; t= e y yt 1 dy / t

0 (2) E[g(x,y)] = g x , y p x , y dxdy

relation with cdf of gamma distribution F a = a ; t
n-Erlang distribution with parameter mean of sums of random variables E[X+Y] = E[X] + E[Y]
the mean of the sample mean is population mean
pdf f x = e x x n1 / n1!, x0
the variance of sample mean is (1/n) * population variance
= 0 , x<0 mean of product of independent random variables
exponential distribution = 1-Erlang distribution with parameter E[g(X)h(Y)] = E[g(x)]E[h(Y)]
relation between Poisson, exponential, gamma distribution covariance of 2 random variables :
Poisson : Cov(X,Y) = E[(X-X)(Y-Y)] = E[XY] = E[X]E[Y]
(1) : number X of successes occurring in n independent trials, Var(X) = Cov(X,X)
each has success probability p, = np if two random variables X,Y are independent, Cov(X,Y)=0
i Cov(aX,bY) = ab Cov(X,Y)
P { X =i }=e n m n m
i! Cov X i , Y j = Cov X i , Y j
(2) : number N with of Poisson events with rate occuring in a fixed i=1 j=1 i=1 j =1
n n n n
time interval of length t
i Var X i= Var X i 2 Cov X i , X j
t
P { N t=i }=e t i=1 i=1 i j
i! mean of the sample total E[To] = n
Exponential : the amount of time X from now until a Poisson event with variance of the sample total Var(To) = n
2

2 2 2

t mean of sample variance is population variance

F t=P {X t }=1e variance of normal sample variance is 24/(n-1)
n-Erlang : the amount of time until n Poisson events with rate has sample mean X and sample variance S2 of normal random sample
occurred. (1) X and S2 are independent
P { X nt }= t ; n (2) X is normal r.v. With (, 2/n)
distribution with n degree of freedom := gamma with =1/2, t=n/2
2 (3) (n-1)S2/2 is a 2 distribution with n-1 degree of freedom
pdf f x =1 /2n/ 2 ex /2 xn / 21 / n/2 , x0 sample covariance of random samples Xi, Yi arising from X, Y
square of unit normal random variable is a gamma with par. (,)
Y 1Y ... X n X
[ X 1 X Y nY ]
S XY = 1
X,Y independent, pdf fx, fy cdf of X+Y n
correlation of random variables X, Y
F X Y a=P { X Y a }= F X a y f Y y dy Cov X ,Y
X ,Y =
Var X Var Y
bivariate normal distribution 1 and 2 are estimators of form an 100(1-)% confidence interval
joint pdf : (1,2) of if P{1<<2} = 1
1
f x , y =
2 x y 1 2

{ [ ]}
2 2
1 x x y y x x y y
exp 2
2
21 x y xy
(1) both the marginal variables X,Y are normal 100(1-)% condidence interval for normal population with known
(2) is the pdf is the correlation of X, Y variance is (L,U) = (X-z/2 * /sqrt(n) , X+z/2 * /sqrt(n) )
(3) X, Y are independent when =0 where P{Z>z} =
[ Z0.025 = 1.96, Z0.05=1.645, Z0.005=2.58 ]
conditional expectation Student's t distribution : Z unit normal, W chi-square with deg=n
discrete : E[X|Y=y] = xP{X=x|Y=y} = x px|y(x|y)
Z
continuous : E[X|Y=y] = -~ xfX|Y (x|y) dx T= is a t distribution with n degree of freedom
E[g(x)|Y=y] = -~ g(x) fX|Y (x|y) dx W /n
E[X] = E[E[X|Y]] = EY[EX[X|Y]] (discrete)
= -~ EX[X|Y=y]fY(y)dy (continuous) n1
n1 /2

Markov's inequality
Chebyshev's inequality
P {Y a }E [Y ]/a , a0
P {W k } 2 /k 2, k0
pdf

mean =0
f t=
2
n n/2
1
t2
n

weak law of large number lim P { X }=1 var = n/(n-2)
n lim n f(t) = 1/sqrt(2) * e^ (-t^2/2)
[proof : take W and k in Chebyshev's to be X and , assume finite ] (n->, t-distribution unit normal distribution)
strong law of large number { = =1
P lim X
n } X1,X2,....,Xn : a random sample from a normal population(,2)
X is the sample mean, S2 is the sample variance
then X - / ( S / sqrt(n)) is a t distribution with n-1 degree of freedom
central limit theorem lim P
n /n{
X
=
} 100(1-)% condidence interval normal population with unknown
variance is (L,U) = (X-t/2,n-1 * S/sqrt(n), X+t/2,n-1 * S/sqrt(n))
where t,n = P{T>t,n}=,

n
T n
lim P o
n{ = }
(the standardized sample mean become the unit normal distribution
Given a large random saple arising from any population distribution, its
100(1-)% condidence interval is
(L,U) = (X-z/2 * S/sqrt(n) , X+z/2 * S/sqrt(n) )
when the sample size become large.) width of interval r = z/2 * S/sqrt(n) => sample size n = (z/2 * s/ r)2
DeMoivre-Laplace limit theorem 100(1-)% condidence interval for a population proportion with success
Sn is the number of success in n independent trials each with success proportion p is
(L,U) = (p - z/2 * sqrt(pq/n), p + z/2 * sqrt(pq/n))
probability p :

sample mean
{
P a
S nnp
np1 p }
b b a sample size n = 1/4 (z/2 /r)2 = p0 (1-p0)(z/2 /r)2
distribution of a function of a random variable
X is a continuous random variable, given g(x), Y=g(X) pdf
linear estimator of population mean (weighted sum of all sample var.) (1) g-1(y)
sample median (2) fY(y) = fX[g-1(y)] * | d/dy g-1(y) |, if y=g(x) for some x
sample midrange (+/2) =0 , otherwise
sample r% trim mean ( r%) exponential random variable with parameter
MSE = E[ (^ - )2 ] = Var(^) - (E[^] - )2 pdf f x =ex mean 1/
estimator 1 is better than 2 if MSE(1) < MSE(2) a
estimator ^ is UNBIASED if E[^] = , or it is BIASED with BIAS E[^] - cdf F a=1e variance 1/ 2
linear estimator with a1+a2+....+an=1 is a linear unbiased estimator of normal random variable
population mean. pdf
sample mean is an unbiased estimator of population mean
sample variance is an unbiased estimator of population variance
population is symmetric and continuous sample median and
trimmed mean are unbiased estimators of population mean expectation and variance of continuous random variable
binomial random variable (n,p)
sample proportion p^ X/ is an unbiased estimator of p
E [ x ] = xf x dx [ ]
Var X =E X 2 E [ X ] 2
minimum-variance estimator ^ has the minimum Var(^) among all
the estimators of
a unbiased estimator ^ wtih minimum var is the best estimator for maximum likelihood estimator
the sample mean is the best unbiased estimator of the population
mean
an estimator is CONSISTENT if lim P{|^-|<} = 1
the sample mean is a consistent estimator of the population mean
[ weak law ]
the variance is a consistent estimator of the population variance
[Chebyshev's ]
moment estimator
(1) m m moment
(2) Mi estimators