El pdf de

Supongamos que $X_1, X_2,...,X_n$ se iid de $N(\mu,\sigma^2)$ con desconocido $\mu \in \mathcal R$ y $\sigma^2>0$

Deje $Z=\frac{X_1-\bar{X}}{S},$ S es la desviación estándar aquí.

Se puede demostrar que $Z$ tiene el pdf de Lebesgue

f (z) = \frac{\sqrt{n} Γ (\frac{n - 1}{2})}{\sqrt{π} (n - 1) Γ (\frac{n - 2}{2})} {[1 - \frac{n z^{2}}{(n - 1)^{2}}]}^{n / 2 - 2} I_{(0, (n - 1) / \sqrt{n})} (| Z |)

$f(z)=\frac{\sqrt{n} \Gamma\left(\frac{n-1}{2}\right)}{\sqrt{\pi}(n-1)\Gamma\left(\frac{n-2}{2}\right)}\left[1-\frac{nz^2}{(n-1)^2}\right]^{n/2-2}I_{(0,(n-1)/\sqrt{n})}(|Z|)$

Mi pregunta es, entonces, ¿cómo obtener este pdf?

La pregunta es de aquí en el ejemplo 3.3.4 para encontrar el UMVUE de $P(X_1 \le c)$ . Puedo entender la lógica y los procedimientos para encontrar el UMVUE pero no sé cómo obtener el pdf.

Creo que esta cuestión también se relacionan con este uno

Muchas gracias por la ayuda o señalar que cualquier referencia relacionada también será apropiada.

self-study umvue Norte profundo
fuente

Respuestas:

Lo que es tan intrigante sobre este resultado es cuánto se parece a la distribución de un coeficiente de correlación. Hay una razón.

Suponga que $(X,Y)$ es bivariada normal con correlación cero y varianza común $\sigma^2$ para ambas variables. Dibuja una muestra iid $(x_1,y_1), \ldots, (x_n,y_n)$ . Es bien sabido y fácilmente establecido geométricamente (como lo hizo Fisher hace un siglo) que la distribución del coeficiente de correlación de la muestra

r = \frac{\sum_{i = 1}^{n} (x_{i} - \bar{x}) (y_{i} - \bar{y})}{(n - 1) S_{x} S_{y}}

$r = \frac{\sum_{i=1}^n(x_i - \bar x)(y_i - \bar y)}{(n-1) S_x S_y}$

f (r) = \frac{1}{B (\frac{1}{2}, \frac{n}{2} - 1)} {(1 - r^{2})}^{n / 2 - 2}, - 1 \leq r \leq 1.

$f(r) = \frac{1}{B\left(\frac{1}{2}, \frac{n}{2}-1\right)}\left(1-r^2\right)^{n/2-2},\ -1 \le r \le 1.$

(Aquí, como de costumbre, y son medias muestrales y y son las raíces cuadradas de los estimadores de varianza insesgados). es la función Beta , para la cual $\bar x$ $\bar y$ $S_x$ $S_y$ $B$

\begin{matrix} (1) & \frac{1}{B (\frac{1}{2}, \frac{n}{2} - 1)} = \frac{Γ (\frac{n - 1}{2})}{Γ (\frac{1}{2}) Γ (\frac{n}{2} - 1)} = \frac{Γ (\frac{n - 1}{2})}{\sqrt{π} Γ (\frac{n}{2} - 1)} . \end{matrix}

$\frac{1}{B\left(\frac{1}{2}, \frac{n}{2}-1\right)} = \frac{\Gamma\left(\frac{n-1}{2}\right)}{\Gamma\left(\frac{1}{2}\right)\Gamma\left(\frac{n}{2}-1\right)} = \frac{\Gamma\left(\frac{n-1}{2}\right)}{\sqrt{\pi}\Gamma\left(\frac{n}{2}-1\right)} . \tag{1}$

Para calcular , podemos explotar su invariancia bajo rotaciones en alrededor de la línea generada por , junto con la invariancia de la distribución de la muestra bajo las mismas rotaciones, y elegir para ser cualquier vector unitario cuyos componentes sumen cero. Uno de esos vectores es proporcional a . Su desviación estándar es $r$ $\mathbb{R}^n$ $(1,1,\ldots, 1)$ $y_i/S_y$ $v = (n-1, -1, \ldots, -1)$

S_{v} = \sqrt{\frac{1}{n - 1} ((n - 1)^{2} + (- 1)^{2} + \dots + (- 1)^{2})} = \sqrt{n} .

$S_v = \sqrt{\frac{1}{n-1}\left((n-1)^2 + (-1)^2 + \cdots + (-1)^2\right)} = \sqrt{n}.$

En consecuencia, debe tener la misma distribución que $r$

\frac{\sum_{i = 1}^{n} (x_{i} - \bar{x}) (v_{i} - \bar{v})}{(n - 1) S_{x} S_{v}} = \frac{(n - 1) x_{1} - x_{2} - \dots - x_{n}}{(n - 1) S_{x} \sqrt{n}} = \frac{n (x_{1} - \bar{x})}{(n - 1) S_{x} \sqrt{n}} = \frac{\sqrt{n}}{n - 1} Z .

$\frac{\sum_{i=1}^n(x_i - \bar x)(v_i - \bar v)}{(n-1) S_x S_v} = \frac{(n-1)x_1 - x_2-\cdots-x_n}{(n-1) S_x \sqrt{n}} = \frac{n(x_1 - \bar x)}{(n-1) S_x \sqrt{n}} = \frac{\sqrt{n}}{n-1}Z.$

Por lo tanto, todo lo que necesitamos es reescalar para encontrar la distribución de : $r$ $Z$

f_{Z} (z) = | \frac{\sqrt{n}}{n - 1} | f (\frac{\sqrt{n}}{n - 1} z) = \frac{1}{B (\frac{1}{2}, \frac{n}{2} - 1)} \frac{\sqrt{n}}{n - 1} {(1 - \frac{n}{(n - 1)^{2}} z^{2})}^{n / 2 - 2}

$f_Z(z) = \big|\frac{\sqrt{n}}{n-1}\big| f\left(\frac{\sqrt{n}}{n-1}z\right) = \frac{1}{B\left(\frac{1}{2}, \frac{n}{2}-1\right)} \frac{\sqrt{n}}{n-1}\left(1- \frac{n}{(n-1)^2}z^2\right)^{n/2-2}$

for $|z| \le \frac{n-1}{\sqrt{n}}$ . Formula (1) shows this is identical to that of the question.

Not entirely convinced? Here is the result of simulating this situation 100,000 times (with $n=4$ , where the distribution is uniform).

The first histogram plots the correlation coefficients of $(x_i,y_i),i=1,\ldots,4$ while the second histogram plots the correlation coefficients of $(x_i,v_i),i=1,\ldots,4)$ for a randomly chosen vector $v_i$ that remains fixed for all iterations. They are both uniform. The QQ-plot on the right confirms these distributions are essentially identical.

Here's the R code that produced the plot.

n <- 4
n.sim <- 1e5
set.seed(17)
par(mfrow=c(1,3))
#
# Simulate spherical bivariate normal samples of size n each.
#
x <- matrix(rnorm(n.sim*n), n)
y <- matrix(rnorm(n.sim*n), n)
#
# Look at the distribution of the correlation of `x` and `y`.
#
sim <- sapply(1:n.sim, function(i) cor(x[,i], y[,i]))
hist(sim)
#
# Specify *any* fixed vector in place of `y`.
#
v <- c(n-1, rep(-1, n-1)) # The case in question
v <- rnorm(n)             # Can use anything you want
#
# Look at the distribution of the correlation of `x` with `v`.
#
sim2 <- sapply(1:n.sim, function(i) cor(x[,i], v))
hist(sim2)
#
# Compare the two distributions.
#
qqplot(sim, sim2, main="QQ Plot")

Reference

R. A. Fisher, Frequency-distribution of the values of the correlation coefficient in samples from an indefinitely large population. Biometrika, 10, 507. See Section 3. (Quoted in Kendall's Advanced Theory of Statistics, 5th Ed., section 16.24.)

whuber
fuente

The link to the reference is broken.

Sextus Empiricus

@Martijn Thank you for checking. I see what you mean--the link works, but it doesn't go to anything relevant! I have fixed it up.

whuber

I'd like to suggest this way to get the pdf of Z by directly calculating the MVUE of $P(X\leq c)$ using Bayes' theorem although it's handful and complex.

Since $E[I_{(-\infty,c)}(X_1)]=P(X_1\leq c)$ and $Z_1=\bar X$ , $Z_2=S^2$ are joint complete sufficient statistic, MVUE of $P(X\leq c)$ would be like this:

ψ (z_{1}, z_{2}) = E [I_{(- \infty, c)} (X_{1}) | z_{1}, z_{2}] = \int_{- \infty}^{\infty} I_{(- \infty, c)} f_{X | Z_{1}, Z_{2}} (x_{1} | z_{1}, z_{2}) d x_{1}

$\psi(z_1,z_2)=E[I_{(-\infty,c)}(X_1)|z_1,z_2]=\int_{-\infty}^{\infty}I_{(-\infty,c)}f_{X|Z_1,Z_2}(x_1|z_1,z_2)dx_1$

Now using Bayes' theorem, we get

f_{X | Z_{1}, Z_{2}} (x_{1} | z_{1}, z_{2}) = \frac{f_{Z_{1}, Z_{2} | X_{1}} (z_{1}, z_{2} | x_{1}) f_{X_{1}} (x_{1})}{f_{Z_{1}, Z_{2}} (z_{1}, z_{2})}

$f_{X|Z_1,Z_2}(x_1|z_1,z_2)={{f_{Z_1,Z_2|X_1}(z_1,z_2|x_1)f_{X_1}(x_1)}\over{f_{Z_1,Z_2}(z_1,z_2)}}$

The denominator $f_{Z_1,Z_2}(z_1,z_2)=f_{Z_1}(z_1)f_{Z_2}(z_2)$ can be written in closed form because $Z_1 \sim N(\mu,\frac{\sigma^2}{n})$ , $Z_2 \sim \Gamma({n-1\over 2},{2 \sigma^2\over n-1})$ are independent of each other.

To get the closed form of numerator, we can adopt these statistics:

W_{1} = \frac{\sum_{i = 2}^{n} X_{i}}{n - 1}

$W_1 = {\sum_{i=2}^n X_i \over n-1}$

W_{2} = \frac{\sum_{i = 2}^{n} X_{i}^{2} - (n - 1) W_{1}^{2}}{(n - 1) - 1}

$W_2 = {\sum_{i=2}^n X_i^2 -(n-1) W_1^2 \over (n-1)-1}$

which is the mean and the sample variance of $X_2, X_3, ..., X_n$ and they are independent of each other and also independent of $X_1$ . We can express these in terms of $Z_1, Z_2$ .

$W_1={n Z_1 - X_1\over n-1}$ , $W_2={(n-1)Z_2+nZ_1^2-X_1^2-(n-1)W_1^2 \over n-2}$

We can use transformation while $X_1=x_1$ ,

f_{Z_{1}, Z_{2} | X_{1}} (z_{1}, z_{2} | x_{1}) = \frac{n}{n - 2} f_{W_{1}, W_{2}} (w_{1}, w_{2}) = \frac{n}{n - 2} f_{W_{1}} (w_{1}) f_{W_{2}} (w_{2})

$f_{Z_1,Z_2|X_1}(z_1,z_2|x_1)={n \over n-2}f_{W_1,W_2}(w_1,w_2)={n \over n-2}f_{W_1}(w_1)f_{W_2}(w_2)$

Since $W_1 \sim N(\mu,\frac{\sigma^2}{n-1})$ , $W_2 \sim \Gamma({n-2\over 2},{2 \sigma^2\over n-2})$ we can get the closed form of this. Note that this holds only for $w_2 \geq 0$ which restricts $x_1$ to $z_1-{n-1 \over \sqrt n}\sqrt{z_2} \leq x_1 \leq z_1+{n-1 \over \sqrt n}\sqrt{z_2}$ .

So put them all together, exponential terms would disappear and you'd get,

f_{X | Z_{1}, Z_{2}} (x_{1} | z_{1}, z_{2}) = \frac{Γ (\frac{n - 1}{2})}{\sqrt{π} Γ (\frac{n - 2}{2})} \frac{\sqrt{n}}{\sqrt{z_{2}} (n - 1)} (1 - {(\frac{\sqrt{n} (x_{1} - z_{1})}{\sqrt{z_{2}} (n - 1)})}^{2})

$f_{X|Z_1,Z_2}(x_1|z_1,z_2)={\Gamma({n-1 \over 2}) \over \sqrt{\pi} \Gamma({n-2 \over 2})} {\sqrt{n} \over \sqrt{z_2} (n-1)} (1-{({\sqrt{n} (x_1 -z_1) \over \sqrt{z_2} (n-1) })}^2)$ where

z_{1} - \frac{n - 1}{\sqrt{n}} \sqrt{z_{2}} \leq x_{1} \leq z_{1} + \frac{n - 1}{\sqrt{n}} \sqrt{z_{2}}

$z_1-{n-1 \over \sqrt n}\sqrt{z_2} \leq x_1 \leq z_1+{n-1 \over \sqrt n}\sqrt{z_2}$ and zero elsewhere.

From this,at this point, we can get the pdf of $Z={X_1- z_1 \over \sqrt{z_2}}$ using transformation.

By the way, the MVUE would be like this :

ψ (z_{1}, z_{2}) = \frac{Γ (\frac{n - 1}{2})}{\sqrt{π} Γ (\frac{n - 2}{2})} \int_{- \frac{π}{2}}^{θ_{c}} c o s^{n - 3} θ d θ

$\psi(z_1,z_2)={\Gamma({n-1 \over 2}) \over \sqrt{\pi} \Gamma({n-2 \over 2})} \int ^{\theta_c} _{-{\pi \over2}} cos^{n-3} \theta d\theta$ while

θ_{c} = s i n^{- 1} (\frac{\sqrt{n} (c - z_{1})}{(n - 1) \sqrt{z_{1}}})

$\theta_c = sin^{-1} ({\sqrt{n}(c-z_1)\over(n-1)\sqrt{z_1}})$ and would be 1 if

c \geq z_{1} + \frac{n - 1}{\sqrt{n} \sqrt{z_{2}}}

$c \geq z_1+{n-1 \over \sqrt{n} \sqrt{z_2} }$

I am not a native English speaker and there could be some awkward sentences. I am studying statistics by myself with text book introduction to mathmatical statistics by Hogg. So there could be some grammatical or mathmatical conceptual mistakes. It would be appreciated if someone correct them.

Thank you for reading.

KDG
fuente