作者yueayase (scrya)
看板Math
标题[机统]求 t- distribution 的 variance
时间Sun May 29 03:44:39 2011
t - distribution 可以用两种方式定义:
(1) Let Z be a standard normal random variable and Y be a chi-square random
variable with v degrees of freedom. Also, Z and Y are independent.
Define a random variable
T = Z / √(Y/v), then the distribution of T is called the t- distribution.
2
(2) The distribution density function Γ((v+1)/2) t -(v+1)/2
f(t) = -------------- (1+---)
√(πv) Γ(v/2) v
-∞ < t < ∞ is called the t- distribution.
2
如果用(1)去算 E(T), E(T ):
1/2 -1/2 -1/2
E(T) = E(Z/(Y/v) ) = v E(Z)E(Y ) = 0 (因为E(Z) = 0)
2 2 2 -1
E(T) = E(Z /(Y/v)) = v E(Z ) E(Y )
2 2 2
因为Z ~ χ , 所以 E(Z ) = 1
1
-1 1
E(Y ) = ------ (用积分算)
v-2
所以 Var(T) = v/(v-2)
但是用(2)的话,
∞
E(T) = ∫ tf(t) = 0 (因为 tf(t) 是奇函数 )
-∞
2 2
但是 E(T ) 好像不好求(Hint是给: Let 1+t / v = 1/u)
但是这样一来,积分上下界不都变为0,好奇怪...
--
※ 发信站: 批踢踢实业坊(ptt.cc)
◆ From: 111.251.162.123
1F:推 Annihilator :f is even... 05/29 04:35
2F:推 ato0715 :因为(1)式中已经假设Z为normal distribution 了 05/29 11:28
3F:→ ato0715 :根据中央极限定理,T分配要变成常数分配的话那样本数 05/29 11:30
4F:→ ato0715 :n要够大(最好能到无限大,但实务上约取30个) 05/29 11:31
5F:→ ato0715 :而在(2)式中,根据积分可得E(T*T)=n/n-2 05/29 11:33
6F:→ ato0715 :如果n很大,则n/(n-2) 趋近於 1 ,(1)(2)式所得相同 05/29 11:34
7F:→ ato0715 :这种解释方法如有错误,还请板上先贤们指教 05/29 11:35