STAT 3690 Lecture 03
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STAT 3690 Lecture 03
2022
Statistical modelling
● What is a statistical model?
一 (Joint) distribution of random variable (RV) of interest
● Rephrase linear regression and logit regression models in terms of distributions
Characterizing/representing univariate distributions
● (scalar-valued) RV X : a real-valued function defined on a sample space Ω
● Cumulative distribution function (cdf): FX (x) = Pr(X s x)
一 right continuous with respect to x
一 non-decreasing w.r.t. x
一 ranging from 0 to 1
● Discrete RV
一 RV X takes countable different values.
一 probability mass function (pmf): pX (x) = Pr(X = x)
● Continuous RV
一 RV X is continuous iff its cdf FX is absolutely continuous with respect to x, i.e., 刁fX , s.t.
x
FX (x) = fX (u)du Ⅴx ∈ 斌.
−∞
一 probability density function (pdf): fX (x) = FX(′) (x).
● Characteristic function
● Moment-generating function
Characterizing/representing joint/multivariate distributions
● Random vector/vector-valued RV 一 艾 = [Xl , . . . , Xp ]干
● Joint cumulative distribution function (joint cdf): Fx (xl , . . . , xp ) = Pr(Xl s xl , . . . , Xp s xp ) 一 right continuous w.r.t. each xi
一 non-decreasing w.r.t. each xi
一 ranging from 0 to 1
● Joint distribution of continuous RVs
一 Joint pdf/density:
fx (xl , . . . , xp ) = Fx (xl , . . . , xp )
一 Multivariate normal (MVN) distribution
● Joint distribution of discrete RVs
一 Joint pmf:
px (xl , . . . , xp ) = Pr(Xl = xl , . . . , Xp = xp )
一 Multinomial distribution
● Multivariate characteristic/moment-generating functions
● Exercise: Suppose that we independently observe an experiment that has m possible outcomes Ol , . . . , Om for n times. Let pl , . . . , pk denote probabilities of Ol , . . . , Om in each experiment re- spectively. Let Xi denote the number of times that outcome Oi occurs in the n repetitions. What is the joint pmf of 艾 = [Xl , . . . , Xm]干 ?
Marginalization
● 艾 = [Xl , . . . , Xp ]干 v = [Xl , . . . , Xq ]干 , and q < p.
● Marginal cdf
FY (xl , . . . , xq ) = x' →∞ lf(i)or(m)all i>q Fx (xl , . . . , xp )
● Marginal pdf of v (when Xl , . . . , Xp are all continous)
∞ ∞
fY (xl , . . . , xq ) = . . . fx (xl , . . . , xp )dxq+l . . . dxp
−∞ −∞
● Marginal pmf of v (when Xl , . . . , Xp are all discrete)
∞ ∞
pY (xl , . . . , xq ) = . . . px (xl , . . . , xp )
xg+l = −∞ xp = −∞
● “marginal” is used to distinguish pdf/pmf of v from the joint pdf/pmf of 艾.
Conditioning = joint/marginal
v = [yl , . . . , yq ]干 and 艾 = [xl , . . . , xp ]干
● Conditional pdf of v given 艾
fY|x(yl , . . . , yq | xl , . . . , xp ) =
● Conditional pmf of v given 艾
pY|x(yl , . . . , yq | xl , . . . , xp ) =
fx ,Y (xl , . . . , xp , yl , . . . , yq )
fx (xl , . . . , xp )
px ,Y (xl , . . . , xp , yl , . . . , yq )
px (xl , . . . , xp )
Transformation of random variables (p-dimentional case)
● Let g = (gl , . . . , gp ): 斌p → 斌p be a one-to-one map with inverse g − l = (gl(−)l , . . . , gl ).
● v = g(艾) and 艾 = g − l (v) are both continuous p-random vectors.
● Jacobian matrix of g − l is 扌 = [∂gl (yl , . . . , yp )/∂yj ]p×p = [∂xi /∂yj ]p×p . 一 |det(扌)| = |det([∂yi /∂xj ]p×p)|− l if replace xj with g − l (yl , . . . , yp )
● fx is known. Then
fY (yl , . . . , yp ) = fx (hl(−)l (yl , . . . , yp ), . . . , hp(−)l (yl , . . . , yp ))|det(扌)|
● Exercise: Let 艾 = [Xl , X2]干 follow the standard bivariate normal, i.e., its pdf is fx (xl , x2 ) = (2π) − l exp{_(xl(2) + x2(2))/2}, (xl , x2 ) ∈ 斌2 .
Find out the joint pdf of v = [Yl , Y2 ]干 , where Yl = ←X l(2) + X2(2) and 0 s Y2 < 2π is angle from the
positive x-axis to the ray from the origin to the point (Xl , X2 ), that is, Y is X in polar co-ordinates.
2022-03-17