Problem 9.2 (4 points)

Let \(Y_1,\dotsc,Y_n\) denote a random sample from a population with mean \(\mu\) and variance \(\sigma^2\). Let \[\begin{align*} \hat{\mu}_1 &= \frac{1}{2}(Y_1+Y_2), \\ \hat{\mu}_2 &= \frac{1}{4}Y_1+\frac{Y_2+\dotsb+Y_{n-1}}{2(n-2)}+\frac{1}{4}Y_n, \\ \hat{\mu}_3 &= \overline{Y}. \end{align*}\]
  1. We will calculate the expected values of \(\hat{\mu}_1\), \(\hat{\mu}_2\), and \(\hat{\mu}_3\). \[\begin{align*} E[\hat{\mu}_1] &= \frac{1}{2}(E[Y_1] + E[Y_2]) = \frac{1}{2}(\mu+\mu)=\mu. \\ E[\hat{\mu}_2] &= \frac{1}{4}E[Y_1]+\frac{1}{2(n-2)}(E[Y_2]+\dotsb+E[Y_{n-1}])+\frac{1}{4}E[Y_n] \\ &= \frac{1}{4}\mu+\frac{1}{2(n-2)}\left(\sum_{i=2}^{n-1}\mu\right)+\frac{1}{4}\mu \\ &= \frac{1}{4}\mu+\frac{1}{2}\mu+\frac{1}{4}\mu \\ &= \mu. \\ E[\hat{\mu}_3] &= E[\overline{Y}] = \mu. \end{align*}\] We see that each of \(\hat{\mu}_1\), \(\hat{\mu}_2\), \(\hat{\mu}_3\) is an unbiased estimator for \(\mu\).
  2. We will calculate the variances of \(\hat{\mu}_1\), \(\hat{\mu}_2\), and \(\hat{\mu}_3\). \[\begin{align*} V[\hat{\mu}_1] &= \frac{1}{4}(V[Y_1] + V[Y_2]) = \frac{1}{4}(\sigma^2+\sigma^2)=\frac{1}{2}\sigma^2. \\ V[\hat{\mu}_2] &= \frac{1}{16}E[Y_1]+\frac{1}{4(n-2)^2}(V[Y_2]+\dotsb+V[Y_{n-1}])+\frac{1}{16}V[Y_n] \\ &= \frac{1}{16}\sigma^2+\frac{1}{4(n-2)^2}\left(\sum_{i=2}^{n-1}\sigma^2\right)+\frac{1}{16}\sigma^2 \\ &= \frac{1}{16}\sigma^2+\frac{1}{4(n-2)}\sigma^2+\frac{1}{16}\sigma^2 \\ &= \frac{n}{8(n-2)} \sigma^2. \\ E[\hat{\mu}_3] &= E[\overline{Y}] = \frac{1}{n} \sigma^2. \end{align*}\] We calculate the relative efficiencies \[\begin{align*} \mathrm{eff}(\hat{\mu}_3,\hat{\mu}_1) &= \frac{V[\hat{\mu}_1]}{V[\hat{\mu}_3]} = \frac{1}{2} n. \\ \mathrm{eff}(\hat{\mu}_3,\hat{\mu}_2) &= \frac{V[\hat{\mu}_2]}{V[\hat{\mu}_3]} = \frac{n^2}{8(n-2)}. \end{align*}\]

Problem 9.4 (2 points)

Let \(Y_1,\dotsc,Y_n\) denote a random sample of size \(n\) from a uniform distribution on the interval \((0 , \theta)\). Let \(Y_{(1)} = \min(Y_1,\dotsc,Y_n)\) and \(Y_{(n)} = \max(Y_1,\dotsc,Y_n)\). Exercise 8.14 implies that \(\hat{\theta}_1 = (n+1)Y_{(1)}\) is an unbiased estimator for \(\theta\), and example 9.1 shows that \(\hat{\theta}_2 = \frac{n+1}{n}Y_{(n)}\) is an unbiased estimator for \(\theta\) with \[ V[\hat{\theta}_2] = \frac{\theta^2}{n(n+2)}. \] Recall that the density for \(Y_{(1)}\) is \[ g_{(1)}(y) = \frac{n}{\theta}\left(1 - \frac{y}{\theta}\right)^{n-1},\quad 0 \le y \le \theta, \] so if we let \(u = y/\theta\) then \(du = \frac{1}{\theta}dy\) and \[ g_{(1)}(y)\,dy = n(1 - u)^{n-1}\,du = \frac{\Gamma(n+1)}{\Gamma(1)\Gamma(n)}u^{1-1}(1-u)^{n-1}\,du. \] Therefore, \(U = \frac{1}{\theta}Y_{(1)}\) has a Beta distribution with parameters \(\alpha=1\), \(\beta=n\), and hence \[\begin{align*} V[\hat{\theta}_1] &= V[(n+1)Y_{(1)}] = V[(n+1)\theta U] \\ &= (n+1)^2 \theta^2 V[U] \\ &= (n+1)^2 \theta^2 \frac{n}{(n+1)^2 (n+2)} \\ &= \frac{n}{n+2} \theta^2. \end{align*}\]

From this we calculate the relative efficiency \[ \textrm{eff}(\hat{\theta}_1,\hat{\theta}_2) = \frac{V[\hat{\theta}_2]}{V[\hat{\theta}_1]} = \frac{\theta^2}{n (n+2)}\cdot \frac{(n+2)}{n \theta^2} = \frac{1}{n^2}. \]

Problem 9.6 (2 points)

Suppose \(Y_1,\dotsc,Y_n\) denotes a random sample of size \(n\) from a Poisson distribution with parameter \(\lambda\). Let \(\hat{\lambda}_1 = \frac{1}{2}(Y_1 + Y_2)\), \(\hat{\lambda}_2 = \overline{Y}\). Then \[\begin{align*} V[\hat{\lambda}_1] &= \frac{1}{4}(V[Y_1] + V[Y_2]) = \frac{1}{2}\lambda. \\ V[\hat{\lambda}_2] &= V[\overline{Y}] = \frac{1}{n}\lambda. \\ \textrm{eff}(\hat{\lambda}_1,\hat{\lambda}_2) &= \frac{V[\hat{\lambda}_2]}{V[\hat{\lambda}_1]} = \frac{\lambda}{n}\cdot \frac{2}{\lambda} = \frac{2}{n}. \end{align*}\]

Problem 9.8 (4 points)

  1.   \[\begin{align*} Y &\sim N(\mu , \sigma^2) \\ f(y) &= \frac{1}{\sigma\sqrt{2\pi}}e^{-\frac{1}{2}\frac{(y-\mu)^2}{\sigma^2}} \\ \ln f(y) &= -\frac{1}{2}(y-\mu)^2 / \sigma^2 - \ln(\sigma \sqrt{2\pi}) \\ -\frac{\partial^2 \ln f}{\partial \mu^2} &= \frac{1}{\sigma^2} \\ E\left[-\frac{\partial^2 \ln f}{\partial \mu^2}(Y)\right] &= \frac{1}{\sigma^2}. \end{align*}\] Therefore, \[ I(\mu) = \frac{1}{n/\sigma^2} = \frac{\sigma^2}{n} = V[\overline{Y}], \] which shows that \(\overline{Y}\) is an efficient estimator for \(\mu\).
  2.   \[\begin{align*} Y &\sim \textrm{Poisson}(\lambda). \\ p(y) &= e^{-\lambda}\lambda^y / y!, y=0,1,2,\dotsc \\ \ln p(y) &= -\lambda + y\ln \lambda - \ln(y!) \\ -\frac{\partial^2 \ln p}{\partial \lambda^2} &= \frac{y}{\lambda^2} \\ E\left[-\frac{\partial^2 \ln p}{\partial \lambda^2}(Y)\right] &= E[Y/\lambda^2] = 1/\lambda. \end{align*}\] Therefore, \[ I(\lambda) = \frac{1}{n/\lambda} = \frac{\lambda}{n} = V[\overline{Y}], \] which shows that \(\overline{Y}\) is an efficient estimator for \(\lambda\).
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