Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Carnegie Mellon University Department of Statistics 36-705 Intermediate Statistics Homework #1 Solutions September 1, 2016 Problem 1: Wasserman 1.3 ∞ Let Ω be a sample space and let A1 , A2 , . . . be events. Define Bn = ⋃∞ i=n Ai and Cn = ⋂i=n Ai . (a) Show that B1 ⊃ B2 ⊃ ⋯ and that C1 ⊂ C2 ⊂ ⋯. Proof. (By Induction) Base. ∞ ∞ i=1 i=2 B1 = ⋃ Ai = A1 ∪ ⋃ Ai = A1 ⋃ B2 Ô⇒ B1 ⊃ B2 ✓ Now for any k ∈ N, ∞ ∞ i=k i=k+1 Bk = ⋃ Ai = Ak ∪ ⋃ Ai = Ak ⋃ Bk+1 Ô⇒ Bk ⊃ Bk+1 . ∎ Proof. (By Induction) Base. ∞ ∞ i=1 i=2 C1 = ⋂ Ai = A1 ∩ ⋂ Ai = A1 ⋂ C2 Ô⇒ C1 ⊂ C2 ✓ Now for any k ∈ N, ∞ ∞ i=k i=k+1 Ck = ⋂ Ai = Ak ∩ ⋂ Ai = Ak ⋂ Ck+1 Ô⇒ Ck ⊂ Ck+1 . ∎ (b) Show that ω ∈ ⋂∞ n=1 Bn if and only if ω belongs to an infinite number of the events A1 , A 2 , . . . . Proof. ∞ “⇐Ô" Suppose ω belongs to an infinite number of the events A1 , A2 , . . . but ω ∉ ⋂∞ n=1 ⋃i=n Ai . Then there exists a j such that ω ∉ ⋃∞ i=j Ai . But this implies ω can be in at most j − 1 of the ∞ Ai , a contradiction. Therefore, ω ∈ ⋂∞ n=1 ⋃i=n Ai . ∞ “Ô⇒" Assume ω ∈ ⋂∞ n=1 ⋃i=n Ai . If ω does not belong to an infinite number of the events A1 , A2 , . . . , then pick the last (according to the index) Ai which contains ω and call it AT . Then ω ∉ Ak , ∀k > T . Hence, ω ∉ ⋂∞ n=1 Bn , a contradiction. Therefore, ω belongs to an infinite number of the events A1 , A2 , . . . . ∎ (c) Show that ω ∈ ⋃∞ n=1 Cn if and only if ω belongs to all the events A1 , A2 , . . . except possibly a finite number of those events. 1 36-705 Intermediate Statistics: Homework 1 Proof. “⇐Ô" Suppose that ω belongs to all the events A1 , A2 , . . . except possibly a finite number of those events. Pick the last Ai that ω is not in and call the next one An0 . ω ∈ An , ∀n ≥ n0 so ∞ ∞ clearly ω ∈ ⋂∞ i=n0 Ai . Since this set is part of the union ⋃n=1 Cn , we’ve shown ω ∈ ⋃n=1 Cn . ∞ “Ô⇒" Suppose ω ∈ ⋃∞ n=1 ⋂i=n Ai . Suppose further that no matter how high we run our index, we can always find an Ai such that ω ∉ Ai . More precisely, for any k ∈ N, ∃j > k such that ∞ ∞ ω ∉ Aj . This of course implies ω ∉ ⋂∞ i=n Ai for any n. That is, ω ∉ ⋃n=1 ⋂i=n Ai , a contradiction. Therefore, ω belongs to all the events A1 , A2 , . . . except possibly a finite number of those events. ∎ Problem 2: Wasserman 1.8 Suppose that P(Ai ) = 1 for each i. Prove that ⎛∞ ⎞ P ⋂ Ai = 1. ⎝ i=1 ⎠ Proof. We will prove the equivalent proposition: ⎛∞ ⎞ P ⋃ Aci = 0. ⎝ i=1 ⎠ In order to show this note that: ∞ ⎛∞ ⎞ ∞ P ⋃ Aci ≤ ∑ P(Aci ) = ∑ 0 = 0. ∎ ⎝ i=1 ⎠ i=1 i=1 Problem 3: Wasserman 1.13 Suppose that a fair coin is tossed repeatedly until both a head and tail have appeared at least once. (a) Describe the sample space Ω. We can partition the sample space into two subsets A = {ω ∶ H⋯H T, n ≥ 1} ´¹¹ ¹ ¸¹ ¹ ¹¶ n times B = {ω ∶ T ⋯T H, n ≥ 1} ² n times Then Ω = A ∪ B. 2 36-705 Intermediate Statistics: Homework 1 (b) What is the probability that three tosses will be required? This can occur in two ways: HHT or TTH. Each has probability ( 12 ) so 3 P (three tosses) = 2 ⋅ 1 1 = . 23 4 Problem 4: Wasserman 2.1 Show that P(X = x) = F (x+ ) − F (x− ). We have by right continuity of the CDF F (x+ ) = F (x) = P(X ≤ x). Also by continuity of probabilities F (x− ) = lim F (x − 1/n) = lim P(X ≤ x − 1/n) = P(X < x). n→∞ n→∞ And these observations together with the fact that P(X = x) = P(X ≤ x) − P(X < x) imply the desired result. Problem 5: Wasserman 2.4 Let X have probability density function ⎧ 1/4 0 < x < 1 ⎪ ⎪ ⎪ ⎪ fX (x) = ⎨3/8 3 < x < 5 ⎪ ⎪ ⎪ ⎪ otherwise. ⎩0 (a) find the cumulative distribution function of X. ⎧ 0 ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ x/4 ⎪ ⎪ ⎪ ⎪ FX (x) = ⎨1/4 ⎪ ⎪ ⎪ 3x−7 ⎪ ⎪ ⎪ 8 ⎪ ⎪ ⎪ ⎪ ⎩1 x<0 0≤x<1 1≤x≤3 . 3<x≤5 x>5 (b) Let Y = 1/X. Find the probability density function fY (y) for Y . Hint: Consider three cases: 51 ≤ y ≤ 13 , 31 ≤ y ≤ 1, and y ≥ 1. 3 36-705 Intermediate Statistics: Homework 1 Following the hint, let us first consider the case when y = 1/x ≥ 1, which occurs exactly when 0 < x ≤ 1. Keep in mind X takes values in the interval (−∞, 0] with probability 0. Now in this interval we have 1 fX (x) = 0<x<1 4 and Y = g(X) = 1/X is indeed monotonic (therefore h(y) = g −1 = 1/y) so we can directly employ the transformation formula for p.d.f.’s. That is, RRR R dh(y) RRRR fY (y) = fX (h(y))RRRRR R= RRR dy RRRRR R R 1 RRRR −1 RRRR 1 R R= , y ≥ 1. 4 RRRR y 2 RRRR 4y 2 R R Using the same method, we first note that 51 ≤ y ≤ 13 corresponds exactly with 3 ≤ x ≤ 5 (we can drop the endpoints since we’re dealing with continuous probability here). For this interval we have 3 fX (x) = 1 < x < 3, 8 and therefore RRR R R R dh(y) RRRR 3 RRRR −1 RRRR 3 1 1 R R RRR = RRR 2 RRR = 2 , fY (y) = fX (h(y))RRR <y< . 5 3 RRR dy RRR 8 RRR y RRR 8y Now for the only remaining interval, to the interval 1 < x < 3 and that Hence, 1 3 < y < 1, we need only note that this corresponds exactly P (X ∈ (1, 3)) = 0. P (Y ∈ (1/3, 1)) = 0. So altogether, ⎧ 3 ⎪ ⎪ 8y 2 ⎪ ⎪ ⎪ 1 fY (y) = ⎨ 4y2 ⎪ ⎪ ⎪ ⎪ ⎪ ⎩0 1 5 <y< y>1 1 3 otherwise. Problem 6: Wasserman 2.7 Let X and Y be independent and suppose that each has a Uniform(0, 1) distribution. Let Z = min{X, Y }. Find the density fZ (z) for Z. Hint: It might be easier to first find P(Z > z). FZ (z) = P(Z ≤ z) = 1 − P(Z > z) = 1 − P(X > z)P(Y > z) = 1 − (1 − z)2 = 2z − z 2 . fZ (z) = FZ′ (z) ⎧ ⎪ ⎪2 − 2z =⎨ ⎪ ⎪ ⎩0 4 0<z<1 otherwise. 36-705 Intermediate Statistics: Homework 1 Problem 7: Wasserman 2.20 Let Z = X − Y . We can easily see that Z ∈ [−1, 1]. To calculate FZ (z) we first consider the case −1 ≤ z ≤ 0: 1+z 1 z2 1 1dydx = +z+ ∫ ∫ 2 2 0 x−z Now for 0 < z ≤ 1, FZ (z) is given by: 1−∫ 1 z ∫ x−z 1dydx = − 0 z2 1 +z+ 2 2 So the density of Z is given by: ⎧ 1+z ⎪ ⎪ ⎪ ⎪ fZ (z) = ⎨1 − z ⎪ ⎪ ⎪ ⎪ ⎩ 0 if −1 ≤ z ≤ 0 if 0 < z ≤ 1 otherwise which is known as triangular distribution. Let W = X/Y , so that W ∈ (0, ∞). We have: P (W ≤ w) = P (X ≤ wY ) = ∫ 1 0 ∫ min(1,wy) 0 1dxdy = ∫ 1 0 min (1, wy) dy Consider now the case w ≤ 1, then the integral becomes: ∫ 0 1 wydy = w 2 Similarly for w > 1 we get: ∫ 1/w 0 wydy + ∫ 1 1/w 1dy = 1 − 1 2w So the density of W is given by: fW ⎧ 1/2 ⎪ ⎪ ⎪ ⎪ (w) = ⎨1/2w2 ⎪ ⎪ ⎪ ⎪ ⎩ 0 if 0 ≤ w ≤ 1 if w > 1 otherwise Problem 8: Wasserman 3.8 Theorem. Let X1 , . . . , Xn be IID and let µ = E(Xi ), σ 2 = V(Xi ). Then E(X n ) = µ, V(X n ) = σ2 and E(Sn2 ) = σ 2 . n Proof. 1 n E(X n ) = E( ∑ Xi ) n i=1 = 1 n ∑ E(Xi ) n i=1 = µ. 5 36-705 Intermediate Statistics: Homework 1 1 n V(X n ) = V( ∑ Xi ) n i=1 E(Sn2 ) = E( = 1 n ∑ V(Xi ) n2 i=1 = σ2 . n 1 n 2 ∑(Xi − X n ) ) n − 1 i=1 = n n 1 ⎛ n 2 ⎞ 2 ∑ E(Xi ) − 2E(X n ∑ Xi ) + E(∑ X n ) n − 1 ⎝ i=1 ⎠ i=1 i=1 = 1 ⎛ 2 2 ⎞ n(σ 2 + µ2 ) − 2nE(X n ) + nE(X n ) n − 1⎝ ⎠ = ⎞ 1 ⎛ n(σ 2 + µ2 ) − n(σ 2 /n + µ2 ) n − 1⎝ ⎠ n−1 2 ⋅σ n−1 = σ2. = Problem 9: Wasserman 3.22 Let X ∼ Uniform(0, 1). Let 0 < a < b < 1. Let ⎧ ⎪ ⎪1 0 < x < b Y =⎨ ⎪ ⎪ ⎩0 otherwise ⎧ ⎪ ⎪1 a < x < 1 Z=⎨ ⎪ ⎪ ⎩0 otherwise and let (a) Are Y and Z independent? Why/Why not? Notice P (Y = 1, Z = 1) = P (X ∈ (a, b)) b−a = 1−0 =b−a and b−0 1−a ⋅ 1−0 1−0 = b(1 − a), P (Y = 1)P (Z = 1) = so Y and Z are not independent. 6 36-705 Intermediate Statistics: Homework 1 (b) Find E(Y ∣Z). Hint: What values z can Z take? Now find E(Y ∣Z = z). E(Y ∣Z = 1) = 1 ⋅ P (Y = 1∣Z = 1) + 0 ⋅ P (Y = 0∣Z = 1) b−a = . 1−a E(Y ∣Z = 0) = 1 ⋅ P (Y = 1∣Z = 0) + 0 ⋅ P (Y = 0∣Z = 0) = 1. Problem 10: Wasserman 3.23 Find the moment generating function for the Poisson, Normal, and Gamma distributions. Poisson ∞ MX (t) = E[etX ] = ∑ etx ⋅ x=0 ∞ t λx −λ (et λ)x e = e−λ ∑ = e−λ+λe x! x! x=0 t −1) MX (t) = eλ(e . Gamma MX (t) = E[etX ] = ∫ ∞ etx 0 ∞ βα β α α−1 −βx α−1 −x(β−t) x e dx = dx. ∫ x e Γ(α) Γ(α) 0 Now we use the property that a p.d.f. integrates to 1 to obtain the formula ∫ ∞ ba a−1 −bx x e dx = 1, Γ(a) 0 which rearranges to ∫ ∞ 0 xa−1 e−bx dx = Γ(a) . ba So our moment generating function is MX (t) = ∞ βα Γ(α) βα α−1 −x(β−t) ⋅ dx = ∫ x e Γ(α) 0 Γ(α) (β − t)α ⎛ β ⎞ , t<β . = ⎝β − t⎠ α Normal First consider X ∼ N (0, 1). MX (t) = E[etX ] = ∫ ∞ −∞ = (compl. the sq.) ∞ x2 x2 etx 1 √ etx− 2 dx = √ ∫ etx− 2 dx 2π 2π −∞ 2 ∞ t2 ∞ ∞ 1 1 u2 2 2 1 et /2 1 Let u = x − t t √ ∫ e 2 e− 2 (x−t) dx = √ ∫ e− 2 (x−t) dx = e 2 ⋅ √ ∫ e− 2 du. 2π −∞ 2π −∞ 2π −∞ 2 7 36-705 Intermediate Statistics: Homework 1 And √ now the integral we are left with is a famous form the the Gaussian integral and is equal to 2π. Therefore, t2 MX (t) = e 2 . Now consider a random variable Y with distribution N (µ, σ 2 ). Using the Central Limit Theorem, we have the equation, Y = µ + σX. We now compute the desired moment generating function. MY (t) = E[eY t ] = E[e(µ+σX)t ] = E[eµt eσX ] = eµt E[eσX ] = eµt MX (σt) = eµt+ 2 σ 1 8 2 t2 .