Partial solutions to assignment 2 Math 5080-1, Spring 2011 p. 283, #1. The question, in mathematical terms, is this: “Suppose X1 , . . . , X20 are an independent sample from N (µ , σ 2 ), where µ = 101 pounds and σ 2 = 4 squared pounds. What is P {X1 + · · · + X20 ≥ 2000}?” Because X1 + · · · + X20 is N (20µ , 20σ 2 ) = N (2020 , 80), we are asked to compute 2000 − 2020 √ P {X1 + · · · X20 ≥ 2000} = 1 − Φ 80 ≈ 1 − Φ(−2.24) = Φ(2.24) ≈ 0.9875, the last computation following from Table 3, p. 603 of your text. p. 283, #2. We need to assume also that S and B are independent [this is assumed tacitly in your textbook, I think]. (a) We want P {S ≥ B} = P {S − B ≥ 0}. But S − B has a N (−0.01 , 0.0013) distribution, being the sum of the two independent normal variables S and −B. Therefore, 0.01 P {S ≥ B} = 1 − Φ √ ≈ 1 − Φ(0.28) ≈ 1 − 0.6103 = 0.3897, 0.0013 thanks to Table 3 on p. 603 of your text. (b) We know instead of before that S ∼ N (1 , σ 2 ) and B ∼ N (1.01 , σ 2 ), and want P {S < B} = 0.95. That is, 0.01 0.95 = P {S − B < 0} = Φ − √ , 2σ 2 since S − B ∼ N (−0.01 , 2σ 2 ). Equivalently (draw a normal-curve picture!), 0.01 0.05 = Φ √ . 2σ 2 Use Table 3 on p. 603 to see that 0.01 √ ≈ 1.65 2σ 2 ⇔ σ= p. 283, #3. 1 0.01 √ ≈ 0.0043. 1.65 × 2 (a) We know that E(X̄) = µ and E(S 2 ) = σ 2 . Therefore, E 2X̄ − 5S 2 = 2µ − 5σ 2 . Now we write things in terms of U and W : X̄ = U , n and owing to (8.2.8) on page 266, S2 = W − n1 U 2 nW − U 2 = . n−1 n(n − 1) Therefore, E 5nW − 5U 2 2U − n n(n − 1) | {z } = 2µ − 5σ 2 . an unbiased estimator of 2µ − 5σ 2 (b) Note that E(X12 ) = · · · = E(Xn2 ) = σ 2 +µ2 . Therefore, E(W ) = n(σ 2 +µ2 ); that is, E(W/n) = σ 2 + µ2 . Consequently, W/n is an unbiased estimator of σ 2 + µ2 . (c) Note that every Yj takes only two values: 1 [with probability P {X1 ≤ c} = FX (c)] and 0 [with probability 1 − FX (c)]. Therefore, E(Yj ) = FX (c). This implies that every Yj is an unbiased estimator of FX (c). A more sensible unbiased estimator can be formed by averaging the Yj ’s: n n X 1 1X E Yj = E(Yj ) = FX (c). n j=1 n j=1 Pn Therefore, n1 j=1 Yj is also an unbiased estimator of FX (c). Although the following is not a part of the problem, it would be good to ask it: Why is the average of Y1 , . . . , Yn a better estimate than any one of them? Check that Var(Y1 ) = · · · = Var(Yn ) = FX (c) {1 − FX (c)} , but n X FX (c){1 − FX (c)} 1 Yj = , Var n j=1 n Pn and this is small when n is large. Therefore, if n is large then n1 j=1 Yj is, with high probability, close to its expectation FX (c). Whereas the probability that 2 Y1 [say] is within units of FX (c) is some fixed amount and cannot be made smaller by taking large samples. p. 283, #4. We compute the joint pdf of (Y1 , Y2 ). First, using calculus variables, if y1 = x1 + x2 and y2 = x1 − x2 , then x1 = y1 + y2 2 and x2 = And the Jacobian is obtained from ∂x1 ∂x1 1 ∂y1 ∂y2 2 J = det = det ∂x 1 ∂x2 2 2 ∂y1 ∂y2 y1 − y2 . 2 1 2 = −1. 2 1 −2 Therefore, fY1 ,Y2 (y1 , y2 ) = fX1 ,X2 (x1 , x2 )|J| 2 2 2 2 1 1 1 =√ e−(x1 −µ) /(2σ ) √ e−(x2 −µ) /(2σ ) 2 2πσ 2πσ 2 2 (x1 − µ) + (x2 − µ) 1 exp − = 4πσ 2 2σ 2 2 x1 − 2(x1 + x2 )µ + x22 + 2µ2 1 exp − = . 4πσ 2 2σ 2 Now we substitute: x21 = y1 + y2 2 2 = y12 + 2y1 y2 + y22 ; 4 x1 + x2 = y1 ; 2 y1 − y2 y 2 − 2y1 y2 + y22 2 x2 = . = 1 2 4 Therefore, x21 − 2(x1 + x2 )µ + x22 + 2µ2 = y12 − 4µy1 + 4µ2 + y22 (y1 − 2µ)2 y2 = + 2. 2 2 2 Plug to find that 1 (y1 − 2µ)2 y22 fY1 ,Y2 (y1 , y2 ) = exp − exp − 2 . 4πσ 2 4σ 2 4σ Therefore, fY1 ,Y2 (y1 , y2 ) can be written as h1 (y1 )h2 (y2 ), which is to say that Y1 and Y2 are independent. Our general theory tells us that Y1 ∼ N (2µ , 2σ 2 ) and Y2 ∼ N (0 , 2σ 2 ) as well [or you can do a little more algebra and obtain this from the previous displayed equation]. 3