1. Suppose that X has a Poisson distribution with mean λ, P(λ), and that Xn is a L sequence of random variables such that Xn −→ X. Is it necessarily true that L (a) Xn I(Xn > 0) −→ X? √ L (b) n(Xn − λ) −→ N (0, λ)? X X e n (c) E 1+e Xn → E e 1+eX ?
(For the “yes” answers, state what theorem are you using.)
2. Let X1 , X2 , . . . be independent random
n variables and that Xj has a uniform distri- bution on the interval (0, 2j). Let Sn = j=1 Xj . (a) Find the mean and variance of Sn. L (b) Show that (Sn − E(Sn ))/ Var(Sn ) −→ N (0, 1), by checking the Lindeberg con- ditions.
3. Consider a multinomial distribution with 6 cells, sample size n, and probability
vector p = 13 (p, p, p, 1 − p, 1 − p, 1 − p) for some unknown probability p, 0 < p < 1. Let nj denote the number of observations that fall in cell j. (a) What is Pearson’s chi-square for testing the hypothesis H0 : p = .5? What is its asymptotic distribution under this hypothesis? (b) What is Hellinger’s chi-square for testing H0 ? (c) What is the approximate large sample distribution of Hellinger’s chi-square if the true value of p is .4?
4. A waiting room contains a long line of n chairs. As customers enter the room, they are seated in the chairs in order of arrival. Male customers appear with probability p and female customers appear with probability 1 − p, independently. Let Sn denote the number of female customers that are seated between two male customers in the line. (a) Find the mean and variance of Sn. (b) Find is the asymptotic (for large n) distribution of Sn properly normalized.
5. Let X1 , X2 , . . . , Xn be a sample from the distribution with distribution function
F (x) = 1 − (1/x) on the interval (1, ∞). (a) Find the pth quantile, xp for this distribution, 0 < p < 1. (b) Find the joint asymptotic distribution of the first and third sample terciles, x̂1/3 and x̂2/3 , properly normalized. Solutions to Midterm Examination Statistics 200C T. Ferguson Fri. May 7, 2010
1. (a) True. From Slutsky’s Theorem, since g(x) = xI(x > 0) is continuous. (b) Very, very false. (c) True. From the Helly-Bray Theorem, since g(x) = exp(x)/(1 + exp(x)) is bounded and continuous.
2. (a) Since Xj has the same distribution as 2jUj where Uj has a uniform distributio 2 2 on (0,1), we have E(Xj ) = j and Var(X n j2 ) = Var(2jUj ) = 4j (1/12) = j /3. So E(Sn ) = n 1 j = n(n + 1)/2 and Var(Sn ) = 1 j /3 = n(n + 1)(2n + 1)/18. (b) Let Xnj = Xj − j, so that EXnj = 0. Then for Bn2 = Var(Sn ), we have Zn /Bn = L (Sn − E(Sn ))/Bn −→ N (0, 1) provided the Lindeberg condition holds. Using Xnj 2 ≤ n2 for all j ≤ n, we have ∞ ∞ 1 1 E[Xnj I(Xnj > Bn )] ≤ 2 2 2 2 2 2 E[Xnj I(n2 > 2 Bn2 )] Bn2 Bn j=1 j=1
= I(n > 2 Bn2 ) = 0
2 for all n sufficiently large
6 3. (a) χ2P = n j=1 ((ni /n) − (1/6))2 /(1/6). Under H0 , χ2P has asymptotically a chi-square distribution with 5 degrees of freedom. 6 (b) χ2H = 4n 1 ( ni /n − 16 )2 . (c) If the true value of p is .4, then for large n, χ2H has approximately a noncentral chi-square distribution with 5 degrees of freedom and noncentrality parameter λ, where we may take ⎡ ⎤ 2 2 2 1 1 1 ⎦ λ = 4 n ⎣3 − +3 − = .04051 n, 15 6 5 6
or 2 2 2 1 1 1 n λ = 6n 3 − +3 − = . 15 6 5 6 25
4. (a) Let Xi = 1 if the ith customer is female and Xi = 0 otherwise. The Xi are i.i.d. Bernoulli with probability of success 1 − p. Let Yi = I(Xi−1 = 0, Xi = 1, Xi+1 = 0 for i = 2, . . . , n − 1. The Yi are 2-dependent and stationary, with EYi = p2 (1 − p), σ00 = p2 (1 − p)− (p2 (1 − p))2 , σ01 = −(p2 (1 − p))2 , and σ02 = p3 (1 − p)2 − (p2 (1 − p))2 . The n−1 number of females seated between two males is Sn = 2 Yi . So ESn = (n − 2)p2 (1 − p) and Var(Sn ) = (n − 2)σ00 + 2(n − 3)σ01 + 2(n − 4)σ02 . √ L (b) n(Sn /(n − 2) − p2 (1 − p)) −→ N (0, σ 2 ), where
