概率统计D 07.01 参数的点估计

$\color{blue}{第七章 参数估计}$

$\color{blue}{\S 7.1 参数的点估计}$

$\color{blue}{一、点估计问题}$

$设总体X的分布函数的形式为已知的F(x, \theta), 其中x是自变量,\\ \theta为位置参数(它可以是一个数,也可以是一个向量).借助于总体\\ X的一个样本(X_1, X_2, \cdots, X_n), 来估计未知参数\theta的值的问\\ 题,称为参数的点估计问题.$

$\color{blue}{二、矩估计法}$

$设总体X的分布函数为F(x, \theta_1, \theta_2, \cdots, \theta_k), 其中\theta_1, \theta_2, \cdots, \theta_k\\ 为k个未知数.假设总体X的各阶原点矩E(X^l)(l = 1, 2, \cdots, k)\\ 存在.则E(X^l)是\theta_1, \theta_2, \cdots, \theta_k的函数,\\ 记作\mu_l = \mu_l(\theta_1, \theta_2, \cdots, \theta_k)即\\ \mu_l(\theta_1, \theta_2, \cdots, \theta_k) = E(X^l)(l = 1, 2, \cdots, k) \\ 对于总体X的样本(X_1, X_2, \cdots, X_n),样本的l阶原点矩为\\ A_l = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i^l, l = 1, 2, \cdots, k \\ 令\\ \mu_l = A_l, l = 1, 2, \cdots, k,$
即
$\left \{ \begin{array}{l} \mu_1(\theta_1, \theta_2, \cdots, \theta_k) = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i, \\ \mu_2(\theta_1, \theta_2, \cdots, \theta_k) = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i^2, \\ \cdots \quad \cdots \quad \cdots \quad \cdots \\ \mu_k(\theta_1, \theta_2, \cdots, \theta_k) = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i^k, \\ \end{array} \right.$

$从上述方程组中解出\theta_1, \theta_2, \cdots, \theta_k, 分别记作\\ \hat{\theta_1} = \hat{\theta_1}(X_1, X_2, \cdots, X_n), \\ \hat{\theta_2} = \hat{\theta_2}(X_1, X_2, \cdots, X_n), \\ \cdots \quad \cdots \quad \cdots \quad \cdots \\ \hat{\theta_k} = \hat{\theta_k}(X_1, X_2, \cdots, X_n), \\ 以此作为未知参数\theta_1, \theta_2, \cdots, \theta_k的估计量,称为矩估计量.$

$如果样本观察值为(x_1, x_2, \cdots, x_n),\\ 则得未知参数\theta_1, \theta_2, \cdots, \theta_k的矩估计值为\\ \hat{\theta_1} = \hat{\theta_1}(x_1, x_2, \cdots, x_n), \\ \hat{\theta_2} = \hat{\theta_2}(x_1, x_2, \cdots, x_n), \\ \cdots \quad \cdots \quad \cdots \quad \cdots \\ \hat{\theta_k} = \hat{\theta_k}(x_1, x_2, \cdots, x_n), \\ 上述估计未知数的方法就叫矩估计法.$

$例1.设总体X服从参数\lambda的泊松分布,其中\lambda > 0为未知,\\ 又设X_1,X_2, \cdots, X_n为X的样本,求\lambda的矩估计量.$
$解: \\ X \sim \pi(\lambda), E(X) = \lambda, 即\mu_1 = E(X) = \lambda,\\ 令\mu_1 = A_1,即\\ \lambda = \dfrac{1}{n} \sum \limits _ {i=1} ^n X_i = \overline X, \\ 得\lambda 的矩估计量为 \hat{\lambda} = \overline X.$

$例2.设总体X服从参数为\lambda的指数分布,其概率密度为$
$f(x) = \left \{ \begin{array}{l} \lambda e^{-\lambda x}, x >0 \\ 0 , x \leq 0 \end{array} \right.$
$其中\lambda > 0为未知,又设X_1, X_2, \cdots, X_n为X的样本,\\ 求\lambda的矩估计量.$
$解:由于\mu_1 = E(X) = \dfrac{1}{\lambda},令 \\ \mu_1= A_1 \\ 即\\ \dfrac{1}{\lambda} = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i = \overline X, \\ 因此得到\lambda的矩估计量为 \hat {\lambda} = \dfrac{1}{\overline X}$

$例3.设总体X在区间[a, b]上服从均匀分布,a与b为未知,\\ X_1, X_2, \cdots, X_n是来自总体X的样本,\\ 求a与b的矩估计量.$
$解:\\ \mu_1 = E(X) = \dfrac{a + b}{2}, \\ \mu_2 = E(X^2) = D(X) + [E(X)]^2 = \dfrac{(b-a)^2}{12} + \dfrac{(a+b)^2}{4}$
$令 \left \{ \begin{array}{l} \mu_1 = A_1, \\ \mu_2 = A_2 \end{array} \right.$
$即\left \{ \begin{array}{l} \dfrac{a + b}{2} = A_1 = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i = \overline X, \\ \dfrac{(b-a)^2}{12} + \dfrac{(a+b)^2}{4} = A_2 = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i^2, \end{array} \right.$
$整理得 \left \{ \begin{array}{l} a + b = 2A_1, \\ b - a = \sqrt{12(A_2 - A_1^2)} \end{array} \right.$
$于是得到a、b的矩估计量为$
$\hat a = A_1 - \sqrt{3(A_2 - A_1^2)} = \overline X - \sqrt{\dfrac{3}{n} \sum \limits _ {i=1} ^n (X_i - \overline X)^2}$
$\hat b = A_1 + \sqrt{3(A_2 - A_1^2)} = \overline X + \sqrt{\dfrac{3}{n} \sum \limits _{i=1} ^n (X_i - \overline X)^2}$

$例4.设总体X的均值为\mu, 方差为\sigma^2,且\sigma > 0, \\ 但\mu 与\sigma均未知,又设总体X的一个样本为(X_1, X_2,\\ \cdots, X_n),求\mu与\sigma^2的矩估计量.$
$解:\mu_1 = E(X) = \mu, \\ \mu_2 = E(X^2) = D(X) + [E(X)]^2 = \sigma^2 + \mu^2.$
$令\left \{ \begin{array}{l} \mu_1 = A_1, \\ \mu_2 = A_2, \end{array} \right.$
$即\left \{ \begin{array}{l} \mu = A_1 = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i = \overline X, \\ \sigma^2 + \mu^2 = A_2 = \dfrac{1}{n} \sum \limits _ {i=1} ^n X_i^2 \end{array} \right.$
$解此方程组得到\mu与\sigma^2的矩估计量为$
$\hat {\mu} = \overline X,$
$\hat {\sigma^2} = A_2 - A_1^2 = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i^2 - \overline X^2 \\ = \dfrac{1}{n} \sum \limits _ {i=1} ^n (X_i - \overline X)^2$
$注:此例说明,无论总体X服从什么分布,样本均值\overline X\\ 都是总体均值\mu的矩估计量,样本二阶中心矩就是总体\\ 方差\sigma^2的矩估计量.$

$例5.某厂生产一批铆钉,现要检验铆钉头部直径,从这批产品中\\ 随机抽取12只,测得头部直径(单位:mm)如下:$
$\begin{array}{l c c } 13.30 & 13.38 & 13.40 & 13.43 & 13.32 & 13.48 & \\ 13.54 & 13.31 & 13.34 & 13.47 & 13.44 & 13.50 & \end{array}$
$设铆钉头部直径这一总体X服从正态分布N(\mu, \sigma^2),\\ 试求\mu与\sigma^2的矩估计量.$
$解:由例4可得\\ \hat {\mu} = \bar x = \dfrac{1}{12}(13.30 + 13.38 + \cdots + 13.50) = 13.41, \\ \hat {\sigma^2} = \dfrac{1}{12} \sum \limits _ {i=1} ^ {12} (x_i - \bar x)^2 \\ = \dfrac{1}{12}[(13.31-13.41)^2 + (13.38 - 13.41)^2 + \cdots + (13.50 - 13.41)^2]$

$\color{blue}{三、极大似然估计法}$

$1.设总体X为离散型随机变量,其分布律为\\ P\lbrace X = x_k \rbrace = p(x_k, \theta), k = 1, 2, \cdots, \\ 其中\theta 为未知参数, 取值范围为\Theta(大写\theta), 设X_1, X_2, \cdots, X_n\\ 为来自X的样本,则X_1, X_2, \cdots, X_n的联合分布律为\\ \prod \limits _{i=1} ^ n p(x_i, \theta).又设x_1, x_2, \cdots, x_n为一组样本值,令\\ L(\theta) = L(x_1, x_2, \cdots, x_n, \theta) = \prod \limits _ {i=1} ^ n p(x_i, \theta), \quad (1) \\ 称L(\theta)为样本的似然函数.$
$若有\hat{\theta} = \hat{\theta}(x_1, x_2, \cdots, x_n) \in \Theta,使得对于一切\theta \in \Theta,有\\ L(\hat{\theta}) \geq L(\theta) \\ 成立,则称\hat{\theta} = \hat{\theta}(x_1, x_2, \cdots, x_n)为\theta的极大(或最大)似然\\ 估计值,相应的统计量\hat{\theta} = \hat{\theta}(X_1, X_2, \cdots, X_n)称为\theta的极\\ 大(或最大)似然估计量.$
$我们规定,使得\dfrac{d L(\theta)}{d \theta} = 0的\hat{\theta}就是\theta的极大似然估计值.由于\\ \ln {x} 是单调增函数,所以L(\theta)与 \ln {L(\theta)}有相同的驻点,因此只需从\\ \dfrac{d \ln{L(\theta)}}{d \theta} = 0 \quad (2) \\ 中解出\hat{\theta}就是\theta的极大似然估计值,称方程(2)为极大似然方程.$

$例6.设总体X \sim \pi(\lambda),X_1, X_2, \cdots, X_n为总体X的样本,\\ 求\lambda的极大似然估计量.$
$解:设样本值为x_1, x_2, \cdots, x_n,由于X的分布律为\\ p(x, \lambda) = \dfrac{\lambda^x}{x!} e^{-\lambda}, x = 1, 2, \cdots, \\ 所以似然函数为\\ L(\lambda) = \prod \limits _ {i=1}^n p(x_ i, \lambda) = e^ {-n \lambda} \prod \limits _ {i=1} ^n \dfrac{\lambda ^{x_i}}{x_i!}, \\ \ln{L(\lambda)} = -n \lambda + \ln{\lambda \sum \limits _ {i=1} ^n x_i }- \sum \limits _{i=1} ^n \ln{(x_i!)} \\ 令 \dfrac{d \ln L(\lambda)}{d \lambda} = -n + \dfrac{1}{\lambda} \sum \limits _ {i=1} ^ n x_i = 0 \\ 得 \lambda 的极大似然估计值为\\ \hat{\lambda} = \dfrac{1}{n} \sum \limits _ {i=1} ^n x_i = \overline x, \\ 因此得到 \lambda 的极大似然估计量为\\ \hat{\lambda} = \dfrac{1}{n} \sum \limits _ {i=1} ^ n X_i = \overline X.$

$例7.设一批产品中含有次品,次品率p未知,从中抽取容量为n的样本,\\ 求p的极大似然估计量.$
$解:从总体中任取一件产品进行观测,其结果可用随机变量X表示如下:$
$X = \left \{ \begin{array}{l} 1,取出一件产品是次品 \\ 0,取出一件产品为正品 \end{array} \right.$
$设X_1, X_2, \cdots, X_n为X的一个样本,观察值为x_1, x_2, \cdots, x_n,则似然数为\\ L(p) = \prod \limits _ {i=1} ^n p(x_i, p) = \prod \limits _ {i=1} ^n p^{x_i}(1-p)^{1-x_i} = p^{\sum \limits _ {i=1} ^n x_i} \cdot (1-p)^{n - \sum \limits _{i=1}^n x_i} \\ \ln{L(p)} = \ln p \cdot \sum \limits _ {i=1} ^ n x_i + \ln (1-p)(n-\sum \limits _ {i=1}^n x_i) \\ 令 \dfrac{d \ln {L(p)} } {dp} = \dfrac{1}{p} \sum \limits _ {i=1} ^n x_i - \dfrac{1}{1-p}(n - \sum \limits _ {i=1} ^n x_i) = 0 \\ 解得p的极大似然估计值为\\ \hat{p} = \dfrac{1}{n} \sum \limits _ {i=1}^n x_i = \bar x \\ 因此p的极大似然估计量为\\ \hat{p} = \dfrac{1}{n} \sum \limits _ {i=1}^n X_i = \overline X$

$2.设总体X为连续型随机变量,其概率密度为f(x, \theta),\theta \in \Theta, \\ \theta为未知参数,设X_1, X_2, \cdots, X_n为来自总体X的样本,其观察\\ 值为x_1, x_2, \cdots, x_n, 则似然函数为\\ L(\theta) = \prod \limits _{i=1} ^n f(x_i, \theta), \quad (3) \\ 似然方程为\\ \dfrac {d}{d \theta} \ln {L(\theta)} = 0 \quad (4) \\ 解出\theta的极大似然估计值为\hat{\theta} = \hat{\theta}(x_1, x_2, \cdots, x_n).极大似然估计量为\\ \hat{\theta} = \hat{\theta}(X_1, X_2, \cdots, X_n).$

$例8.设总体X的密度为$
$f(x, \lambda) = \left \{ \begin{array}{l} \lambda e^{-\lambda x}, x > 0, \\ 0, x \leq 0 \end{array} \right.$
$其中\lambda > 0为未知参数,（X_1, X_2, \cdots, X_n)为样本,\\ 求\lambda 的极大似然估计量.$
$解:设样本值为(x_1, x_2, \cdots, x_n)(x_i > 0, i = 1, 2, \cdots, n), 似然函数为\\ L(\lambda) = \prod \limits _ {i=1}^n f(x_i, \lambda) = \prod \limits _ {i=1} ^n \lambda e^{-\lambda x_i} = \lambda ^n e^{-\lambda \sum \limits _ {i=1}^n x_i} \\ \ln {L(\lambda)} = n \ln \lambda - \lambda \sum \limits _{i=1} ^n x_i \\ 令 \dfrac{d}{d \lambda} {\ln L(\lambda)} = \dfrac{n}{\lambda} - \sum \limits _{i=1}^n x_i = 0 \\ 得极大似然估计值为\\ \hat{\lambda} = \dfrac{n}{\sum \limits _{i=1}^n x_i} = \dfrac{1}{\overline x} \\ 于是得到\lambda 的极大似然估计量为\hat{\lambda} = \dfrac{1}{\overline X}$

$例9.设总体X的概率密度为$
$f(x) = \left \{ \begin{array}{l} \theta x^{\theta -1}, 0 < x < 1 \\ 0, 其它 \end{array} \right.$
$又设X_1, X_2, \cdots, X_n为X的样本,\\ 求\theta的矩估计量与极大似然估计量.$
$解:(1)由于\\ \mu_1 = EX = \int_{-\infty}^{+\infty} xf(x) dx = \int_0^1 x \theta x^{\theta -1} dx = \dfrac{\theta}{\theta + 1}, \\ 令\mu_1 = A_1, \\ 即\dfrac{\theta}{\theta + 1} = \dfrac{1}{n} \sum \limits _{i=1}^n X_i = \overline X, \\ 解得\theta 的矩估计量为 \hat{\theta} = \dfrac{\overline X}{1 - \overline X} \\ (2) 设样本值为x_1, x_2, \cdots, x_n (0 < x_i < 1),似然函数为\\ L(\theta) = \prod \limits _ {i=1}^n \theta x_i^{\theta - 1} = \theta^n(\prod \limits _{i=1}^n x_i)^{\theta - 1} \\ \ln {L(\theta)} = n \ln \theta + (\theta - 1) \sum \limits _ {i=1} ^n \ln x_i \\ 令 \dfrac{d}{d \theta} {\ln L(\theta)} = \dfrac{n}{\theta} + \sum \limits _{i=1}^n \ln x_i = 0, \\ 解得\theta 的极大似然估计值为\\ \hat{\theta} = -\dfrac{n}{\sum \limits _{i=1}^n \ln x_i } \\ 因此,\theta的极大似然估计量为\\ \hat{\theta} = -\dfrac{n}{\sum \limits _{i=1}^n \ln X_i }$

$3.设总体X的分布中含有k个参数\theta_1, \theta_2, \cdots, \theta_k,\\ 则似然函数是这些未知参数的函数\\ L = L(\theta_1, \theta_2, \cdots, \theta_k), \\ 取对数后,求出 \ln L关于\theta_i的偏导数并令它等于0,\\ 得到似然方程组\\ \dfrac{\partial \ln L}{\partial \theta_i} = 0, i = 1, 2, \cdots, k \\ 由此方程组解得\theta_i的极大似然估计值\hat {\theta_i}.$

$例10.设总体X \sim N(\mu, \sigma^2),\mu与\sigma^2未知,\\ （X_1,X_2, \cdots, X_n)为总体X的样本,\\ 求\mu与\sigma^2的极大似然估计量.$
$解:X的概率密度为\\ f(x, \mu, \sigma^2) = \dfrac{1}{\sqrt{2 \pi} \sigma} e^{-\frac{(x - \mu)^2}{2 \sigma^2}}, -\infty < x < +\infty \\ 设x_1, x_2, \cdots, x_n为样本值,似然函数为\\ L(\mu, \sigma^2) = \prod \limits _ {i=1}^n f(x_i, \mu, \sigma^2) = \prod \limits _ {i=1}^n \dfrac{1}{\sqrt{2 \pi} \sigma} e^{-\frac{(x_i - \mu)^2}{2 \sigma^2}} \\ = (2 \pi)^{-\frac{n}{2}} \dfrac{1}{\sigma^n} e^{-\frac{1}{2\sigma^2} \sum \limits _{i=1}^n (x_i - \mu)^2 } \\ \ln{L(\mu, \sigma^2)} = -\dfrac{n}{2} \ln{2 \pi} - \dfrac{n}{2} \ln {\sigma^2} - \dfrac{1}{2 \sigma^2} \sum \limits _{i=1}^n (x_i - \mu)^2$

$例11.设总体X在区间[a, b]上服从均匀分布,\\ 其中a、b未知,X_1, X_2, \cdots, X_n为总体X的样本,\\ 求a、b的极大似然估计量.$
$解:X的概率密度为$
$f(x) = \left \{ \begin{array}{l} \dfrac{1}{b-a}, a \leq x \leq b, \\ 0, 其它 \end{array} \right.$
$设样本值为x_1, x_2, \cdots, x_n(a \leq x_i \leq b),似然函数为\\ L(a, b) = \dfrac{1}{(b-a)^n} \\ 因为L(a, b) 是a的单调增函数,a越大,L(a, b)就越大,但a不能大于\\ x_{(1)} = \min \lbrace x_1, x_2, \cdots, x_n \rbrace; 又因为L(a, b)是b的单调减函数,\\ b越小,L(a, b)就越大,但b不能小于x_{(n)} = max \lbrace x_1, x_2, \cdots, x_n \rbrace.\\ 对于满足a \leq x_{(1)}, b \geq x_{(n)}的任意a,b有 \\ L(a, b) = \dfrac{1}{(b-a)^n} \leq \dfrac{1}{[x_{(n)} - x_{(1)}]^n}$
$当a= x_{(1)}, b = x_{(n)}时,L(a, b)取得最大值\dfrac{1}{[x_{(n)} - x_{(1)}]^n}$
$所以a, b的极大似然估计值为$
$\hat{a} = x_{(1)} = \min \lbrace x_1, x_2, \cdots, x_n \rbrace,$
$\hat{b} = x_{(n)} = \max \lbrace x_1, x_2, \cdots, x_n \rbrace,$
$a,b的极大似然估计量为$
$\hat{a} = X_{(1)} = \min \lbrace X_1, X_2, \cdots, X_n \rbrace,$
$\hat{b} = X_{(n)} = \max \lbrace X_1, X_2, \cdots, X_n \rbrace,$

$4.极大似然估计的性质$
$设\mu(\theta)是关于未知参数\theta的函数,\theta \in \Theta, \mu(\theta)具有单值反函数,\\ 又设\hat{\theta}是总体分布中所含参数\theta的极大似然估计,则\hat{\mu} = \mu(\hat \theta) \\ 是\mu的极大似然估计.$

$\color{blue}{四、估计量的评选标准}$

1.无偏性
估计量是样本的函数,它是一个随机变量,由不同的方法得到的估计量可能相同也可能不同.而对同一估计量,由不同的样本观察值得到的估计值也可能不同.我们很自然地要求估计量的期望等于参数的真值,即无偏性.

$定义:设\hat{\theta} = \hat{\theta}(X_1, X_2, \cdots, X_n)是未知参数\theta的估计量,\\ 若E(\hat \theta) = \theta,则称\hat{\theta}为\theta的无偏估计(量).$

$例12.设(X_1, X_2, \cdots, X_n)是来自具有有限均值\mu与方差\sigma^2的\\ 总体X的一个样本.证明:样本均值\overline X是\mu的无偏估计,样本方差\\ S^2是\sigma^2的无偏估计.$
$证:\hat \mu = \overline X. \\ \hat {\sigma^2} = S^2 = \dfrac{1}{n-1} \sum \limits _ {i=1} ^n (X_i - \overline X)^2 = \dfrac{1}{n-1}[\sum \limits _ {i=1}^n X_i^2 - n \overline X^2 ] \\ E(\hat \mu) = E(\dfrac{1}{n} \sum \limits _ {i=1}^n X_i) = \dfrac{1}{n} \sum \limits _ {i=1} ^n E(X_i) = \dfrac{1}{n} \sum \limits _ {i=1} ^n \mu = \mu \\ E(\hat {\sigma^2}) = E(\dfrac{1}{n-1} [\sum \limits _ {i=1}^n X_i^2 - n \overline X^2]) \\ = \dfrac{1}{n-1} [\sum \limits E(X_i^2) - nE(\overline X^2)] \\ = \dfrac{1}{n-1}[nE(X^2) - nE(\overline X^2)] \\ = \dfrac{n}{n-1}[D(X) + (E(X))^2 - D(\overline X) - (E(\overline X))^2] \\ = \dfrac{n}{n-1}[\sigma^2 + \mu^2 - \dfrac{\sigma^2}{n} - \mu^2] \\ = \sigma^2 \\ 因此,\hat{\mu} = \overline X与\hat {\sigma^2} = S^2 分别为\mu 与 \sigma^2的无偏估计.$

$例13.设总体X的均值为\mu(X_1, X_2, X_3)是总体X的样本,\\ 证明下列两个估计量$
$\hat \mu_1 = X_2, \hat \mu_2 = \dfrac{1}{2}X_1 + \dfrac{1}{6}X_2 + \dfrac{1}{3}X_3 \\ 都是\mu的无偏估计.$
$解:由于 E(\hat{\mu_1}) = E(X_2) = E(X) = \mu, \\ E(\hat{\mu_2}) = \dfrac{1}{2}E(X_1) + \dfrac{1}{6}E(X_2) + \dfrac{1}{3}E(X_3) \\ = \dfrac{1}{2} \mu + \dfrac{1}{6} \mu + \dfrac{1}{3} \mu \ = \mu \\ 所以\hat{\mu_1} 与 \hat{\mu_2}都是\mu的无偏估计. \\ (只需k_1 + k_2 + \cdots + k_n = 1,则\hat{\mu} = k_1X_1 + k_2X_2 + \cdots + k_nX_n\\ 就是\mu的无偏估计)$

2.有效性
$设\hat{\theta_1}与\hat{\theta_2}是参数\theta的两个无偏估计量,若D(\hat{\theta_1}) < D(\hat{\theta_2}), 则称\hat{\theta_1}比\hat{\theta_2}有效.$

$例14.比较例13中\hat{\mu_1}与\hat{\mu_2}哪个更有效.$

3.一致性
$设\hat{\theta} = \hat{\theta}(X_1, X_2, \cdots, X_n)为参数\theta的估计量,\\ 若当n \rightarrow \infty时, \hat{\theta}按概率收敛于\theta, 即对于任意正数 \varepsilon, 有\\ \lim \limits _ {n \rightarrow \infty} P\lbrace |\hat{\theta} - \theta| < \varepsilon \rbrace = 1, \\ 则称\hat{\theta} 为\theta的一致估计(量)$